Introduction to Machine Learning with Python – Winter 2023/24

Machine learning is reaching notable success when solving complex tasks in many fields. This course serves as in introduction to basic machine learning concepts and techniques, focusing both on the theoretical foundation, and on implementation and utilization of machine learning algorithms in Python programming language. High attention is paid to the ability of application of the machine learning techniques on practical tasks, in which the students try to devise a solution with highest performance.

Python programming skills are required, together with basic probability theory knowledge.

About

Official name: Introduction to Machine Learning with Python
SIS code: NPFL129
Semester: winter
E-credits: 5
Examination: 2/2 C+Ex
Instructors: Jindřich Libovický (lecture), Zdeněk Kasner, Tomáš Musil (practicals), Milan Straka (assignments & ReCodEx), Petr Kašpárek, Marek Seltenhofer, Matej Straka (teaching assistants)

This course is also part of the inter-university programme prg.ai Minor. It pools the best of AI education in Prague to provide students with a deeper and broader insight into the field of artificial intelligence. More information is available at prg.ai/minor.

Timespace Coordinates

lecture: Czech lecture is held on Tuesday 10:40 in S5, English lecture on Tuesday 17:20 in S9; first lecture is on Oct 03
practicals: Czech practicals are held on Tuesday 12:20 in S5, English practicals on Wednesday 9:00 in S9; first practicals are on Oct 03

All lectures and practicals will be recorded and available on this website.

Course Objectives

After this course students should…

Be able to reason about task/problems suitable for ML
- Know when to use classification, regression and clustering
- Be able to choose from this method Linear and Logistic Regression, Multilayer Perceptron, Nearest Neighbors, Naive Bayes, Gradient Boosted Decision Trees, $k$ -means clustering
Think about learning as (mostly probabilistic) optimization on training data
- Know how the ML methods learn including theoretical explanation
Know how to properly evaluate ML
- Think about generalization (and avoiding overfitting)
- Be able to choose a suitable evaluation metric
- Responsibly decide what model is better
Be able to implement ML algorithms on a conceptual level
Be able to use Scikit-learn to solve ML problems in Python

Lectures

1. Introduction to Machine Learning Slides PDF Slides CS Lecture CS Practicals EN Practicals EN Practicals Slides linear_regression_manual linear_regression_features Questions

2. Linear Regression, SGD Slides PDF Slides CS Lecture CS Practicals EN Practicals linear_regression_l2 linear_regression_sgd feature_engineering rental_competition Questions

3. Peceptron, Logistic Regression Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals perceptron logistic_regression_sgd grid_search thyroid_competition Questions

4. Multiclass Logistic Regression, Multilayer Perceptron Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals (part) softmax_classification_sgd mlp_classification_sgd mnist_competition Questions

5. MLP, Softmax as MaxEnt classifier, F1 score Slides PDF Slides multilabel_classification_sgd diacritization CS Lecture CS Practicals EN Lecture EN Practicals Questions

6. Representing Text (TF-IDF, Word2Vec) Slides PDF Slides tf_idf imdb_sentiment diacritization_dictionary CS Lecture CS Practicals EN Lecture EN Practicals Questions

7. K Nearest Neighbors, Naive Bayes Slides PDF Slides k_nearest_neighbors naive_bayes isnt_it_ironic CS Lecture EN Lecture EN Practicals Questions

8. Correlation, Model Combination Slides PDF Slides metric_correlation miniaturization CS Lecture CS Practicals EN Lecture EN Practicals Questions

9. Decision Trees, Random Forests Slides PDF Slides decision_tree random_forest human_activity_recognition CS Lecture CS Practicals EN Lecture 1/2 EN Lecture 2/2 EN Practicals Questions

10. Gradient Boosted Decision Trees Slides PDF Slides gradient_boosting nli_competition CS Lecture CS Practicals EN Lecture EN Practicals Questions

11. SVD, PCA, k-means Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals pca kmeans Questions

12. Statistical Hypothesis Testing, Model Comparison Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals bootstrap_resampling permutation_test Questions

13. Machine Learning Ethics, Final Summary Slides PDF Slides CS Lecture EN Lecture EN Practicals Questions

License

Unless otherwise stated, teaching materials for this course are available under CC BY-SA 4.0.

The lecture content, including references to some additional study materials. The main study material is the Pattern Recognition and Machine Learning by Christopher Bishop, referred to as PRML.

Note that the topics in italics are not required for the exam.

1. Introduction to Machine Learning

Oct 03 Slides PDF Slides CS Lecture CS Practicals EN Practicals EN Practicals Slides linear_regression_manual linear_regression_features Questions

Learning objectives. After the lecture you shoud be able to…

Explain to an non-expert what machine learning is.
Explain the difference between classification and regression.
Implement a simple linear-algebra-based algorithm for training linear regression.

Covered topics and where to find more:

Introduction to machine learning
Basic definitions [Sections 1 and 1.1 of PRML]
Linear regression model [Section 3.1 of PRML]

After the lecture: short and non-comprehensive recap quiz.

2. Linear Regression, SGD

Oct 10 Slides PDF Slides CS Lecture CS Practicals EN Practicals linear_regression_l2 linear_regression_sgd feature_engineering rental_competition Questions

Learning objectives. After the lecture you shoud be able to

Reason about overfitting in terms of model capacity.
Use $L^2$ -regularization to control model capacity.
Explain what the difference between parameters and hyperparameters is.
Tell what the basic probability concepts are (joint, marginal, conditional probability; expected value, mean, variance).
Mathematically describe and implement the stochastic gradient descent algorithm.
Use both numerical and categorical features in linear regression.

Covered topics and where to find more:

L2 regularization in linear regression [Section 1.1, 3.1.4 of PRML]
Random variables and probability distributions [Section 1.2, 1.2.1 of PRML]
Expectation and variance [Section 1.2.2 of PRML]
Gradient descent [Section 5.2.4 of PRML]
- Stochastic gradient descent solution of linear regression [slides]
Linear regression demo by Jared Willber

After the lecture: short and non-comprehensive recap quiz.

3. Peceptron, Logistic Regression

Oct 17 Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals perceptron logistic_regression_sgd grid_search thyroid_competition Questions

Learning objectives. After the lecture you shoud be able to

Think about binary classification using geometric intuition and use the perceptron algorithm.
Define the main concepts of information theory (entropy, cross-entropy, KL-divergence) and prove their basic properties.
Derive training objectives using the maximum likelihood principle.
Implement and use logistic regression for binary classification with SGD.

Covered topics and where to find more:

Linear models for classification [Section 4.1.1 of PRML]
Perceptron algorithm [Section 4.1.7 of PRML]
Probability distributions [Bernoulli Section 2.1, Categorical Section 2.2, Gaussian Section 2.3 of PRML]
Information theory [Section 1.6 of PRML]
Maximum likelihood estimation [Section 1.2.5 of PRML]
Logistic regression [Section 4.3.2 of PRML]
Cross-validation [Section 1.3 of PRML]
Logistic regression demo by Jared Willber

After the lecture: short and non-comprehensive recap quiz.

4. Multiclass Logistic Regression, Multilayer Perceptron

Oct 24 Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals (part) softmax_classification_sgd mlp_classification_sgd mnist_competition Questions

Learning objectives. After the lecture you shoud be able to

Implement muticlass classification with softmax.
Reason about linear regression, logistic regression and softmax classification in a single probablistic framework: with different target distributions, activation functions and training using maximum likelihood estimate.
Explain multi-layer perceptron as a further generalization of linear models.

Covered topics and where to find more:

Generalized linear models
MSE as MLE [Section 3.1.1 of PRML]
Multiclass logistic regression [Section 4.3.4 of PRML]
Multilayer perceptron (neural network) [Sections 5-5.3 of PRML]
- Interactive demo of MLP by Andrej Karpathy
Universal approximation theorem

After the lecture: short and non-comprehensive recap quiz.

5. MLP, Softmax as MaxEnt classifier, F1 score

Oct 31 Slides PDF Slides multilabel_classification_sgd diacritization CS Lecture CS Practicals EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Implement training of multi-layer perceptron using SGD.
Explain the theoretical foundation behind the softmax activation function (including the necessary math).
Choose a suitable evaluation metric for various classification tasks.

Covered topics and where to find more:

Lagrange multipliers [Appendix E of PRML]
- A Simple Explanation of Why Lagrange Multipliers Works, a blog post by Andrew Chamberlain
Derivation of softmax via the maximum entropy principle [The equivalence of logistic regression and maximum entropy models writeup]
$F_1$ score and $F_β$ score
K-nearest neighbors [Section 2.5.2 of PRML]

6. Representing Text (TF-IDF, Word2Vec)

Nov 7 Slides PDF Slides tf_idf imdb_sentiment diacritization_dictionary CS Lecture CS Practicals EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Use TF-IDF for representing documents and explain its information-theoretical interpretation.
Explain training of Word2Vec as a special case of logistic regression.
Use pre-trained word embeddings for simple NLP tasks.

Covered topics and where to find more:

TF-IDF
Word2Vec (original paper Mikolov et al., 2013)

After the lecture: short and non-comprehensive recap quiz.

7. K Nearest Neighbors, Naive Bayes

Nov 14 Slides PDF Slides k_nearest_neighbors naive_bayes isnt_it_ironic CS Lecture EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Implement and use $k$ -nearest neighbors for classification and regression.
Explain the very basic principles of Bayesian thinking.
Implement and use Naive Bayes Classifier.

Covered topics and where to find more:

K-nearest neighbors [Section 2.5.2 of PRML]
Naive Bayes classifier [Basic idea in Section 8.2.2 of PRML]

After the lecture: short and non-comprehensive recap quiz.

8. Correlation, Model Combination

Nov 21 Slides PDF Slides metric_correlation miniaturization CS Lecture CS Practicals EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Explain and implement different ways of measuring correlation: Pearson's correlation, Spearman's correlation, Kendall's $\tau$ .
Decide if correlation is a good metric for your model.
Measure inter-annotator agreement and draw conclusions for data cleaning and for limits of your models.
Use correlation with human judgment to validate evaluation metrics.
Ensemble models with uncorrelated predictions.
Distill ensembles into smaller models.

Covered topics and where to find more:

Covariance and correlation
Inter-annotator agreement
Model ensembling [Section 14.2 of PRML]

After the lecture: short and non-comprehensive recap quiz.

9. Decision Trees, Random Forests

Nov 28 Slides PDF Slides decision_tree random_forest human_activity_recognition CS Lecture CS Practicals EN Lecture 1/2 EN Lecture 2/2 EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Implement Decision Trees and Random Forests for classification and regression
Explain how the splitting criterion depend on optimized loss function
Tell how Random Forests differ from Gradient Boosted Decision Trees

Covered topics and where to find more:

Decision trees [Section 14.4 of PRML]
- Decision trees demo by Jared Wilber & Lucía Santamaría
Random forests
- Random forets demo by Jenny Yeon & Jared Wilber

After the lecture: short and non-comprehensive recap quiz.

10. Gradient Boosted Decision Trees

Dec 05 Slides PDF Slides gradient_boosting nli_competition CS Lecture CS Practicals EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Explain second-order optimization methods
Implement gradient boosted decision trees for regression and classification
Decide what supervised machine learning approach is suitable for particular problems

Covered topics and where to find more:

Gradient boosting decision trees [Paper XGBoost: A Scalable Tree Boosting System]
- Interactive Playground by Alex Rogozhnikov

After the lecture: short and non-comprehensive recap quiz.

11. SVD, PCA, k-means

Dec 12 Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals pca kmeans Questions

Learning objectives. After the lecture you shoud be able to

Theoretically explain Singular Value Decomposition (SVD), prove it exists and explain what the Eckart-Young theorem says.
Theoretically explain Principal Component Analysis (PCA) and say how it explains the variance in the data based on SVD.
Use SVD or PCA for dimensionality reduction, data visualization and data whitening.
Implement the $k$ -means algorithm and use it for clustering.

Covered topics and where to find more:

Singular value decomposition [Gilbert Strang: Linear Algebra and Learning from Data. Wellesley- Cambridge Press, 2019. Chapters I.8, I.9.]
- Lectures 4-7 in the corresponding course from MIT Courseware
Principal component analysis [Sections 12.1 and 12.4.2 of PRML]
Power iteration algorithm
K-Means clustering [Section 9.1 of PRML]

After the lecture: short and non-comprehensive recap quiz.

12. Statistical Hypothesis Testing, Model Comparison

Dec 19 Slides PDF Slides CS Lecture CS Practicals EN Lecture EN Practicals bootstrap_resampling permutation_test Questions

Learning objectives. After the lecture you shoud be able to

Explain foundations of statistical hypothesis testing.
Reason about multiple comparison problem.
Use Bootstrap Resampling and Permutation Tests to compare machine learning models.

Covered topics and where to find more:

Statistical hypothesis testing
Bootstrap resampling
Paired bootstrap test
Random permutation test

13. Machine Learning Ethics, Final Summary

Jan 09 Slides PDF Slides CS Lecture EN Lecture EN Practicals Questions

Learning objectives. After the lecture you shoud be able to

Explain what main theoretical ethical frameworks are.
Reason about ethical problems in various stages of developing ML System.

Covered topics and where to find more:

Main ethical frameworks: deontology, utilitarism
Ethical problem examples
- Problem definition (some takes are inherently problematic)
- Data collection (biases, unethical acquisition)
- Modeling problems (bias amplification via discretization, privacy violation via memorization)
- Evaluation problems (misleading metrics, unexpected consequences of optimization)
- Deployment (train-test mismatch, feedback loops)
More materials:
- Rachel Thomas: 11 Short Machine Learning Ethics Videos
- Mark Coeckelbergh: AI Ethics / Etika umělé inteligence

Requirements

To pass the practicals, you need to obtain at least 70 points, excluding the bonus points. Note that up to 40 points above 70 (both bonus and non-bonus) will be transfered to the exam. In total, assignments for at least 105 points (not including the bonus points) will be available.

Environment

The tasks are evaluated automatically using the ReCodEx Code Examiner.

The evaluation is performed using Python 3.11, scikit-learn 1.3.0, numpy 1.24.2, scipy 1.11.2, pandas 2.1.0, and matplotlib 3.7.2. You should install the exact version of these packages yourselves.

Teamwork

Solving assignments in teams (of size at most 3) is encouraged, but everyone has to participate (it is forbidden not to work on an assignment and then submit a solution created by other team members). All members of the team must submit in ReCodEx individually, but can have exactly the same sources/models/results. Each such solution must explicitly list all members of the team to allow plagiarism detection using this template.

No Cheating

Cheating is strictly prohibited and any student found cheating will be punished. The punishment can involve failing the whole course, or, in grave cases, being expelled from the faculty. While discussing assignments with any classmate is fine, each team must complete the assignments themselves, without using code they did not write (unless explicitly allowed). Of course, inside a team you are allowed to share code and submit identical solutions.

linear_regression_manual

Deadline: Oct 17, 7:59 a.m. 3 points

Starting with the linear_regression_manual.py template, solve a linear regression problem using the algorithm from the lecture which explicitly computes the matrix inversion. Then compute root mean square error on the test set.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 linear_regression_manual.py --test_size=0.1

52.38

python3 linear_regression_manual.py --test_size=0.5

54.58

python3 linear_regression_manual.py --test_size=0.9

59.46

linear_regression_features

Deadline: Oct 17, 7:59 a.m. 3 points

Starting with the linear_regression_features.py template, use scikit-learn to train a model of a 1D curve.

Try using a concatenation of features $x^1, x^2, …, x^D$ for $D$ from 1 to a given range, and report RMSE of every such configuration.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 linear_regression_features.py --data_size=10 --test_size=5 --range=6

Maximum feature order 1: 0.74 RMSE
Maximum feature order 2: 1.87 RMSE
Maximum feature order 3: 0.53 RMSE
Maximum feature order 4: 4.52 RMSE
Maximum feature order 5: 1.70 RMSE
Maximum feature order 6: 2.82 RMSE

python3 linear_regression_features.py --data_size=30 --test_size=20 --range=9

Maximum feature order 1: 0.56 RMSE
Maximum feature order 2: 1.53 RMSE
Maximum feature order 3: 1.10 RMSE
Maximum feature order 4: 0.28 RMSE
Maximum feature order 5: 1.60 RMSE
Maximum feature order 6: 3.09 RMSE
Maximum feature order 7: 3.92 RMSE
Maximum feature order 8: 65.11 RMSE
Maximum feature order 9: 3886.97 RMSE

python3 linear_regression_features.py --data_size=50 --test_size=40 --range=9

Maximum feature order 1: 0.63 RMSE
Maximum feature order 2: 0.73 RMSE
Maximum feature order 3: 0.31 RMSE
Maximum feature order 4: 0.26 RMSE
Maximum feature order 5: 1.22 RMSE
Maximum feature order 6: 0.69 RMSE
Maximum feature order 7: 2.39 RMSE
Maximum feature order 8: 7.28 RMSE
Maximum feature order 9: 201.70 RMSE

linear_regression_l2

Deadline: Oct 24, 7:59 a.m. 2 points

Starting with the linear_regression_l2.py template, use scikit-learn to train L2-regularized linear regression models and print the results of the best of them.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 linear_regression_l2.py --test_size=0.15

0.49 52.11

python3 linear_regression_l2.py --test_size=0.80

0.10 53.53

linear_regression_sgd

Deadline: Oct 24, 7:59 a.m. 4 points

Starting with the linear_regression_sgd.py, implement minibatch SGD for linear regression and compare the results to an explicit linear regression solver.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 linear_regression_sgd.py --batch_size=10 --epochs=50 --learning_rate=0.01

Test RMSE: SGD 114.118, explicit 115.6
Learned weights: 6.864 6.907 -1.208 -2.252 -1.464 -13.323 13.909 4.883 -11.468 -0.229 37.803 -5.191 ...

python3 linear_regression_sgd.py --batch_size=10 --epochs=50 --learning_rate=0.1

Test RMSE: SGD 111.395, explicit 115.6
Learned weights: 11.559 12.428 -1.529 -2.236 -1.575 -8.868 18.842 3.882 -7.175 -1.373 38.918 -6.522 ...

python3 linear_regression_sgd.py --batch_size=10 --epochs=50 --learning_rate=0.001

Test RMSE: SGD 151.210, explicit 115.6
Learned weights: 1.885 -0.580 -0.386 0.389 -1.745 -6.994 6.787 3.019 -8.013 0.353 15.712 -3.322 ...

python3 linear_regression_sgd.py --batch_size=1 --epochs=50 --learning_rate=0.01

Test RMSE: SGD 111.395, explicit 115.6
Learned weights: 11.559 12.429 -1.529 -2.236 -1.574 -8.868 18.843 3.882 -7.174 -1.373 38.917 -6.522 ...

python3 linear_regression_sgd.py --batch_size=50 --epochs=50 --learning_rate=0.01

Test RMSE: SGD 136.015, explicit 115.6
Learned weights: 2.940 0.504 -0.555 0.143 -2.088 -10.664 9.146 4.607 -11.620 0.129 24.294 -4.089 ...

python3 linear_regression_sgd.py --batch_size=50 --epochs=500 --learning_rate=0.01

Test RMSE: SGD 111.914, explicit 115.6
Learned weights: 9.360 9.428 -1.333 -2.646 -1.379 -11.248 16.352 4.153 -9.041 -0.755 38.872 -5.881 ...

python3 linear_regression_sgd.py --batch_size=50 --epochs=500 --learning_rate=0.01 --l2=0.1

Test RMSE: SGD 113.521, explicit 115.6
Learned weights: 8.013 7.818 -1.227 -2.234 -1.491 -11.592 14.863 4.343 -9.807 -0.575 36.745 -5.487 ...

feature_engineering

Deadline: Oct 24, 7:59 a.m. 3 points

Starting with the feature_engineering.py template, learn how to perform basic feature engineering using scikit-learn.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 feature_engineering.py --dataset=diabetes

-0.5745 -0.9514 1.797 -0.4984 0.4751 0.9487 -0.6961 0.7574 0.06019 1.625 0.33 0.5465 -1.033 0.2863 -0.2729 -0.545 0.3999 -0.4351 -0.03458 -0.9334 0.9052 -1.71 0.4742 -0.452 -0.9026 0.6623 -0.7206 -0.05727 -1.546 3.23 -0.8959 0.8539 1.705 -1.251 1.361 0.1082 2.92 0.2484 -0.2368 -0.4729 0.347 -0.3775 -0.03 -0.8099 0.2257 0.4507 -0.3307 0.3598 0.0286 0.7719 0.9 -0.6604 0.7185 0.0571 1.541 0.4845 -0.5272 -0.0419 -1.131 0.5736 0.04559 1.231 0.003623 0.0978 2.64
0.2776 -0.9514 0.08366 -1.148 -1.592 -1.397 -0.4687 -0.7816 -0.3766 -1.973 0.07706 -0.2641 0.02322 -0.3186 -0.442 -0.3878 -0.1301 -0.217 -0.1045 -0.5477 0.9052 -0.0796 1.092 1.515 1.329 0.4459 0.7436 0.3583 1.877 0.007 -0.09602 -0.1332 -0.1169 -0.03921 -0.06539 -0.03151 -0.1651 1.317 1.827 1.603 0.5379 0.8971 0.4322 2.264 2.535 2.224 0.7462 1.245 0.5996 3.141 1.952 0.6548 1.092 0.5261 2.757 0.2197 0.3663 0.1765 0.9247 0.6109 0.2944 1.542 0.1418 0.7431 3.893
0.8198 1.051 -0.683 -0.8108 -0.6896 -0.4871 -0.2413 -0.03186 -0.2682 0.04527 0.6721 0.8617 -0.5599 -0.6647 -0.5653 -0.3993 -0.1978 -0.02612 -0.2199 0.03711 1.105 -0.7179 -0.8522 -0.7248 -0.512 -0.2536 -0.03348 -0.2819 0.04758 0.4665 0.5538 0.471 0.3327 0.1648 0.02176 0.1832 -0.03092 0.6574 0.5591 0.3949 0.1956 0.02583 0.2175 -0.0367 0.4755 0.3359 0.1664 0.02197 0.185 -0.03121 0.2373 0.1175 0.01552 0.1306 -0.02205 0.05822 0.007686 0.06472 -0.01092 0.001015 0.008544 -0.001442 0.07194 -0.01214 0.002049
0.9747 1.051 1.211 0.6803 0.6207 -0.9859 -1.151 1.547 2.783 2.853 0.9501 1.025 1.18 0.6631 0.605 -0.961 -1.122 1.508 2.712 2.781 1.105 1.273 0.715 0.6524 -1.036 -1.21 1.626 2.925 2.999 1.467 0.8239 0.7517 -1.194 -1.394 1.873 3.37 3.456 0.4628 0.4222 -0.6707 -0.7829 1.052 1.893 1.941 0.3852 -0.6119 -0.7143 0.96 1.727 1.771 0.972 1.135 -1.525 -2.743 -2.813 1.325 -1.78 -3.203 -3.284 2.392 4.304 4.413 7.743 7.94 8.142
-0.1872 -0.9514 0.1739 -1.171 -0.5149 -0.8915 0.5925 -0.8211 0.3554 -0.1302 0.03503 0.1781 -0.03254 0.2193 0.09637 0.1669 -0.1109 0.1537 -0.06651 0.02438 0.9052 -0.1654 1.115 0.4899 0.8482 -0.5637 0.7812 -0.3381 0.1239 0.03023 -0.2037 -0.08952 -0.155 0.103 -0.1428 0.06178 -0.02264 1.372 0.6032 1.044 -0.6941 0.9619 -0.4163 0.1526 0.2651 0.459 -0.3051 0.4228 -0.183 0.06706 0.7948 -0.5282 0.732 -0.3168 0.1161 0.3511 -0.4865 0.2106 -0.07717 0.6742 -0.2918 0.1069 0.1263 -0.04628 0.01696
0.9747 -0.9514 -0.1869 -0.3058 2.659 2.728 0.3651 0.7574 0.676 -0.1302 0.9501 -0.9274 -0.1822 -0.2981 2.592 2.659 0.3559 0.7382 0.6589 -0.1269 0.9052 0.1778 0.291 -2.53 -2.596 -0.3474 -0.7206 -0.6431 0.1239 0.03494 0.05717 -0.497 -0.51 -0.06824 -0.1416 -0.1264 0.02434 0.09354 -0.8132 -0.8344 -0.1117 -0.2316 -0.2067 0.03983 7.07 7.254 0.9708 2.014 1.797 -0.3463 7.443 0.9961 2.066 1.844 -0.3553 0.1333 0.2765 0.2468 -0.04755 0.5736 0.512 -0.09864 0.457 -0.08804 0.01696
1.982 -0.9514 0.715 0.4877 -0.5149 -0.3253 -0.01389 -0.8211 -0.4683 -0.4813 3.927 -1.885 1.417 0.9664 -1.02 -0.6447 -0.02753 -1.627 -0.928 -0.9537 0.9052 -0.6803 -0.464 0.4899 0.3095 0.01322 0.7812 0.4455 0.4579 0.5113 0.3487 -0.3682 -0.2326 -0.009932 -0.5871 -0.3348 -0.3441 0.2378 -0.2511 -0.1586 -0.006774 -0.4004 -0.2284 -0.2347 0.2651 0.1675 0.007152 0.4228 0.2411 0.2478 0.1058 0.004519 0.2671 0.1523 0.1566 0.000193 0.01141 0.006505 0.006685 0.6742 0.3845 0.3952 0.2193 0.2254 0.2316
1.362 1.051 -0.1418 -0.2337 2.193 1.084 1.123 -0.03186 1.76 -0.3935 1.855 1.432 -0.1932 -0.3183 2.987 1.476 1.53 -0.04339 2.397 -0.536 1.105 -0.1491 -0.2456 2.305 1.139 1.18 -0.03348 1.85 -0.4136 0.02011 0.03314 -0.311 -0.1537 -0.1593 0.004518 -0.2496 0.05581 0.05462 -0.5125 -0.2532 -0.2625 0.007445 -0.4113 0.09196 4.809 2.376 2.463 -0.06986 3.86 -0.8629 1.174 1.217 -0.03452 1.907 -0.4264 1.261 -0.03578 1.977 -0.4419 0.001015 -0.05607 0.01253 3.098 -0.6926 0.1548
2.059 -0.9514 1.031 1.69 1.174 0.8206 -1.606 3.046 2.055 1.274 4.24 -1.959 2.122 3.48 2.417 1.69 -3.306 6.273 4.231 2.623 0.9052 -0.9806 -1.608 -1.117 -0.7807 1.528 -2.898 -1.955 -1.212 1.062 1.742 1.21 0.8458 -1.655 3.14 2.118 1.313 2.857 1.984 1.387 -2.714 5.149 3.473 2.153 1.378 0.9633 -1.885 3.576 2.412 1.495 0.6734 -1.318 2.5 1.686 1.045 2.578 -4.891 -3.3 -2.045 9.279 6.26 3.88 4.223 2.618 1.623
0.2776 1.051 -0.48 -0.0173 0.8245 1.178 -0.1655 0.7574 -0.1065 -0.218 0.07706 0.2918 -0.1333 -0.004801 0.2289 0.327 -0.04594 0.2102 -0.02956 -0.06051 1.105 -0.5046 -0.01818 0.8666 1.238 -0.1739 0.7961 -0.1119 -0.2291 0.2304 0.008303 -0.3958 -0.5654 0.07944 -0.3636 0.05111 0.1046 0.0002992 -0.01426 -0.02037 0.002862 -0.0131 0.001842 0.00377 0.6798 0.9711 -0.1364 0.6245 -0.08779 -0.1797 1.387 -0.1949 0.8921 -0.1254 -0.2568 0.02739 -0.1253 0.01762 0.03608 0.5736 -0.08064 -0.1651 0.01134 0.02321 0.04752

python3 feature_engineering.py --dataset=linnerud

1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...

python3 feature_engineering.py --dataset=wine

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.2976 1.31 0.1177 0.9155 0.7783 0.5271 0.4232 0.6668 -1.122 1.048 0.7913 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1.282 -1.41 0.8239 -0.4697 -0.1538 0.2001 -1.152 1.377 -0.9146 -0.6609 0.7232 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1.442 1.419 0.5492 -1.8 1.03 0.9984 -1.302 0.8977 -0.02794 -0.2336 1.336 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.4121 -0.9989 -1.844 -0.3312 -1.263 -0.6176 -0.6272 -0.3989 -1.174 0.4073 0.301 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.9155 -0.7526 -0.2354 0.8601 -0.7751 -0.3001 0.4232 -0.02594 -1.174 1.646 -0.3936 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.778 0.6071 0.7454 -1.245 0.586 0.9888 -1.527 0.1517 -0.02794 0.06549 1.105 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1.43 1.465 0.2746 -0.2204 0.8079 0.6233 -0.5521 -0.5766 0.03261 -0.3191 1.064 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.03446 0.3424 1.295 0.3614 -1.13 -1.445 1.173 -1.465 -0.2442 -0.7464 -0.3255 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0.8809 -0.8529 1.295 0.777 1.03 1.2 -0.6272 1.431 0.2316 1.048 0.2193 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.6752 -1.154 -1.765 -0.02646 -0.2869 -0.001945 -0.7772 -0.9496 -0.2096 0.7491 1.268 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...

rental_competition

Deadline: Oct 24, 7:59 a.m. 3 points+4 bonus

This assignment is a competition task. Your goal is to perform regression on the data from a bike rental shop. The train set contains 1000 instances, each instance consists of 12 features, both integral and real.

The rental_competition.py template shows how to load the training data, downloading it if needed. Furthermore, it shows how to save a trained estimator and how to load it during prediction.

The performance of your system is measured using root mean squared error and your goal is to achieve RMSE less than 100. Note that you can use any number of generalized linear models from sklearn to solve this assignment (but no decision trees, MLPs, …).

perceptron

Deadline: Oct 31, 7:59 a.m. 2 points

Starting with the perceptron.py template, implement the perceptron algorithm.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 perceptron.py --data_size=100 --seed=17

Learned weights 4.10 2.94 -1.00

python3 perceptron.py --data_size=50 --seed=320

Learned weights -2.30 -1.96 -2.00

python3 perceptron.py --data_size=200 --seed=92

Learned weights 4.43 1.54 -2.00

logistic_regression_sgd

Deadline: Oct 31, 7:59 a.m. 5 points

Starting with the logistic_regression_sgd.py, implement minibatch SGD for logistic regression.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 logistic_regression_sgd.py --data_size=100 --batch_size=10 --epochs=9 --learning_rate=0.5

After epoch 1: train loss 0.3259 acc 94.0%, test loss 0.3301 acc 96.0%
After epoch 2: train loss 0.2321 acc 96.0%, test loss 0.2385 acc 98.0%
After epoch 3: train loss 0.1877 acc 98.0%, test loss 0.1949 acc 98.0%
After epoch 4: train loss 0.1612 acc 98.0%, test loss 0.1689 acc 98.0%
After epoch 5: train loss 0.1435 acc 98.0%, test loss 0.1517 acc 98.0%
After epoch 6: train loss 0.1307 acc 98.0%, test loss 0.1396 acc 98.0%
After epoch 7: train loss 0.1208 acc 98.0%, test loss 0.1304 acc 96.0%
After epoch 8: train loss 0.1129 acc 98.0%, test loss 0.1230 acc 96.0%
After epoch 9: train loss 0.1065 acc 98.0%, test loss 0.1170 acc 96.0%
Learned weights 2.77 -0.60 0.12

python3 logistic_regression_sgd.py --data_size=95 --test_size=45 --batch_size=5 --epochs=9 --learning_rate=0.5

After epoch 1: train loss 0.2429 acc 96.0%, test loss 0.3187 acc 93.3%
After epoch 2: train loss 0.1853 acc 96.0%, test loss 0.2724 acc 93.3%
After epoch 3: train loss 0.1590 acc 96.0%, test loss 0.2525 acc 93.3%
After epoch 4: train loss 0.1428 acc 96.0%, test loss 0.2411 acc 93.3%
After epoch 5: train loss 0.1313 acc 98.0%, test loss 0.2335 acc 93.3%
After epoch 6: train loss 0.1225 acc 96.0%, test loss 0.2258 acc 93.3%
After epoch 7: train loss 0.1159 acc 96.0%, test loss 0.2220 acc 93.3%
After epoch 8: train loss 0.1105 acc 96.0%, test loss 0.2187 acc 93.3%
After epoch 9: train loss 0.1061 acc 96.0%, test loss 0.2163 acc 93.3%
Learned weights -0.61 3.61 0.12

python3 logistic_regression_sgd.py --data_size=95 --test_size=45 --batch_size=1 --epochs=9 --learning_rate=0.7

After epoch 1: train loss 0.1141 acc 96.0%, test loss 0.2268 acc 93.3%
After epoch 2: train loss 0.0867 acc 96.0%, test loss 0.2150 acc 91.1%
After epoch 3: train loss 0.0797 acc 98.0%, test loss 0.2320 acc 88.9%
After epoch 4: train loss 0.0753 acc 96.0%, test loss 0.2224 acc 88.9%
After epoch 5: train loss 0.0692 acc 96.0%, test loss 0.2154 acc 88.9%
After epoch 6: train loss 0.0749 acc 98.0%, test loss 0.2458 acc 88.9%
After epoch 7: train loss 0.0638 acc 96.0%, test loss 0.2190 acc 88.9%
After epoch 8: train loss 0.0644 acc 98.0%, test loss 0.2341 acc 88.9%
After epoch 9: train loss 0.0663 acc 98.0%, test loss 0.2490 acc 88.9%
Learned weights -1.07 7.33 -0.40

grid_search

Deadline: Oct 31, 7:59 a.m. 2 points

Starting with grid_search.py template, perform a hyperparameter grid search, evaluating hyperparameter performance using a stratified k-fold crossvalidation, and finally evaluate a model trained with best hyperparameters on all training data. The easiest way is to utilize sklearn.model_selection.GridSearchCV.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 grid_search.py --test_size=0.5

Test accuracy: 98.33%

python3 grid_search.py --test_size=0.7

Test accuracy: 96.98%

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 grid_search.py --test_size=0.5

Rank: 11 Cross-val: 86.7% lr__C: 0.01  lr__solver: lbfgs polynomial__degree: 1
Rank:  5 Cross-val: 92.7% lr__C: 0.01  lr__solver: lbfgs polynomial__degree: 2
Rank: 11 Cross-val: 86.7% lr__C: 0.01  lr__solver: sag   polynomial__degree: 1
Rank:  5 Cross-val: 92.7% lr__C: 0.01  lr__solver: sag   polynomial__degree: 2
Rank:  7 Cross-val: 90.8% lr__C: 1     lr__solver: lbfgs polynomial__degree: 1
Rank:  3 Cross-val: 96.8% lr__C: 1     lr__solver: lbfgs polynomial__degree: 2
Rank:  7 Cross-val: 90.8% lr__C: 1     lr__solver: sag   polynomial__degree: 1
Rank:  4 Cross-val: 96.8% lr__C: 1     lr__solver: sag   polynomial__degree: 2
Rank: 10 Cross-val: 90.1% lr__C: 100   lr__solver: lbfgs polynomial__degree: 1
Rank:  1 Cross-val: 97.2% lr__C: 100   lr__solver: lbfgs polynomial__degree: 2
Rank:  9 Cross-val: 90.5% lr__C: 100   lr__solver: sag   polynomial__degree: 1
Rank:  2 Cross-val: 97.0% lr__C: 100   lr__solver: sag   polynomial__degree: 2
Test accuracy: 98.33%

python3 grid_search.py --test_size=0.7

Rank: 11 Cross-val: 87.9% lr__C: 0.01  lr__solver: lbfgs polynomial__degree: 1
Rank:  5 Cross-val: 91.8% lr__C: 0.01  lr__solver: lbfgs polynomial__degree: 2
Rank: 11 Cross-val: 87.9% lr__C: 0.01  lr__solver: sag   polynomial__degree: 1
Rank:  5 Cross-val: 91.8% lr__C: 0.01  lr__solver: sag   polynomial__degree: 2
Rank:  7 Cross-val: 91.3% lr__C: 1     lr__solver: lbfgs polynomial__degree: 1
Rank:  3 Cross-val: 95.9% lr__C: 1     lr__solver: lbfgs polynomial__degree: 2
Rank:  7 Cross-val: 91.3% lr__C: 1     lr__solver: sag   polynomial__degree: 1
Rank:  4 Cross-val: 95.7% lr__C: 1     lr__solver: sag   polynomial__degree: 2
Rank: 10 Cross-val: 89.2% lr__C: 100   lr__solver: lbfgs polynomial__degree: 1
Rank:  1 Cross-val: 96.5% lr__C: 100   lr__solver: lbfgs polynomial__degree: 2
Rank:  9 Cross-val: 89.2% lr__C: 100   lr__solver: sag   polynomial__degree: 1
Rank:  2 Cross-val: 96.1% lr__C: 100   lr__solver: sag   polynomial__degree: 2
Test accuracy: 96.98%

thyroid_competition

Deadline: Oct 31, 7:59 a.m. 3 points+4 bonus

This assignment is a competition task. Your goal is to perform binary classification – given medical data with 15 binary and 6 real-valued attributes, predict whether thyroid is functioning normally or not. The train set and test set consist of ~3.5k instances.

The thyroid_competition.py template shows how to load training data, downloading it if needed. Furthermore, it shows how to save a trained estimator and how to load it during prediction.

The performance of your system is measured using accuracy of correctly predicted examples and your goal is to achieve at least 96% accuracy. Note that you can use any number of generalized linear models from sklearn to solve this assignment (but no decision trees, MLPs, …).

softmax_classification_sgd

Deadline: Nov 7, 7:59 a.m. 3 points

Starting with the softmax_classification_sgd.py, implement minibatch SGD for multinomial logistic regression.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 softmax_classification_sgd.py --batch_size=10 --epochs=2 --learning_rate=0.005

After epoch 1: train loss 0.3130 acc 90.8%, test loss 0.3529 acc 88.7%
After epoch 2: train loss 0.2134 acc 93.9%, test loss 0.2450 acc 92.5%
Learned weights:
  -0.03 -0.10 0.01 0.06 -0.07 0.04 -0.05 0.05 0.07 -0.10 ...
  0.09 0.08 -0.12 -0.08 -0.10 0.09 -0.03 -0.06 0.02 -0.01 ...
  0.05 0.07 0.01 -0.03 -0.05 0.06 0.04 -0.10 -0.03 0.08 ...
  0.02 -0.05 -0.01 0.10 0.11 0.09 -0.05 0.06 -0.09 0.04 ...
  -0.07 -0.07 -0.10 -0.01 -0.06 -0.07 -0.08 0.04 -0.04 0.01 ...
  -0.07 -0.05 0.14 0.06 0.02 0.14 0.05 0.04 -0.04 0.03 ...
  -0.09 -0.04 -0.11 -0.06 -0.04 -0.10 -0.09 0.05 0.05 -0.01 ...
  0.07 0.01 0.02 -0.04 0.04 -0.01 0.11 -0.06 0.03 -0.03 ...
  0.02 -0.02 0.01 -0.03 0.00 -0.03 -0.09 -0.03 0.08 -0.07 ...
  0.04 -0.04 -0.05 0.05 -0.04 -0.05 0.09 -0.08 -0.01 -0.04 ...

python3 softmax_classification_sgd.py --batch_size=1 --epochs=1 --learning_rate=0.005 --test_size=1597

After epoch 1: train loss 1.7683 acc 73.5%, test loss 2.0028 acc 72.2%
Learned weights:
  -0.03 -0.10 0.03 0.08 0.03 0.03 -0.07 0.05 0.07 -0.15 ...
  0.09 0.08 -0.25 -0.15 -0.17 0.11 -0.00 -0.06 0.02 -0.05 ...
  0.05 0.06 0.07 0.04 -0.12 0.11 0.07 -0.10 -0.03 0.16 ...
  0.02 -0.03 0.03 0.14 0.03 0.08 -0.09 0.06 -0.09 0.09 ...
  -0.07 -0.08 -0.22 -0.07 -0.11 -0.27 -0.13 0.04 -0.04 -0.00 ...
  -0.07 -0.08 0.17 0.16 0.17 0.39 0.07 0.04 -0.03 0.03 ...
  -0.09 -0.04 -0.13 -0.10 -0.03 -0.16 -0.09 0.05 0.05 -0.03 ...
  0.07 0.02 0.10 0.03 0.09 -0.05 0.13 -0.08 0.03 -0.05 ...
  0.02 0.00 0.02 -0.17 -0.01 -0.04 -0.12 -0.03 0.08 -0.09 ...
  0.04 -0.04 -0.02 0.06 -0.07 -0.05 0.16 -0.08 -0.01 -0.01 ...

python3 softmax_classification_sgd.py --batch_size=100 --epochs=3 --learning_rate=0.05

After epoch 1: train loss 4.2661 acc 77.8%, test loss 4.4413 acc 75.5%
After epoch 2: train loss 0.4290 acc 90.5%, test loss 0.5414 acc 89.8%
After epoch 3: train loss 0.6189 acc 88.0%, test loss 0.5752 acc 89.2%
Learned weights:
  -0.03 -0.10 -0.04 0.08 -0.07 -0.02 -0.05 0.05 0.07 -0.12 ...
  0.09 0.06 -0.23 -0.08 -0.12 0.11 -0.04 -0.06 0.02 -0.09 ...
  0.05 0.09 0.07 -0.01 -0.08 0.01 0.02 -0.10 -0.03 0.16 ...
  0.02 -0.04 0.03 0.15 0.18 0.09 -0.09 0.06 -0.09 0.10 ...
  -0.07 -0.07 -0.14 -0.07 -0.07 -0.13 -0.11 0.03 -0.04 -0.01 ...
  -0.07 -0.03 0.28 0.07 0.06 0.29 0.11 0.04 -0.03 0.08 ...
  -0.09 -0.04 -0.16 -0.09 -0.04 -0.13 -0.09 0.05 0.04 -0.05 ...
  0.07 0.01 0.06 -0.01 0.05 0.06 0.18 -0.04 0.03 -0.03 ...
  0.02 -0.03 -0.03 -0.08 -0.00 -0.03 -0.12 -0.04 0.08 -0.12 ...
  0.04 -0.05 -0.04 0.06 -0.09 -0.09 0.12 -0.09 -0.01 -0.03 ...

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 softmax_classification_sgd.py --batch_size=10 --epochs=10 --learning_rate=0.005

After epoch 1: train loss 0.3130 acc 90.8%, test loss 0.3529 acc 88.7%
After epoch 2: train loss 0.2134 acc 93.9%, test loss 0.2450 acc 92.5%
After epoch 3: train loss 0.1366 acc 96.8%, test loss 0.1735 acc 94.6%
After epoch 4: train loss 0.1374 acc 96.2%, test loss 0.1705 acc 94.0%
After epoch 5: train loss 0.1169 acc 97.2%, test loss 0.1667 acc 95.1%
After epoch 6: train loss 0.0978 acc 97.5%, test loss 0.1340 acc 96.1%
After epoch 7: train loss 0.0878 acc 98.0%, test loss 0.1366 acc 95.9%
After epoch 8: train loss 0.0889 acc 97.5%, test loss 0.1515 acc 95.1%
After epoch 9: train loss 0.0819 acc 98.0%, test loss 0.1336 acc 96.5%
After epoch 10: train loss 0.0801 acc 97.9%, test loss 0.1342 acc 96.4%
Learned weights:
  -0.03 -0.10 0.01 0.08 -0.05 0.01 -0.06 0.05 0.07 -0.10 ...
  0.09 0.07 -0.15 -0.02 -0.21 0.13 -0.01 -0.06 0.02 -0.07 ...
  0.05 0.08 0.01 -0.03 -0.05 0.06 0.04 -0.10 -0.03 0.09 ...
  0.02 -0.03 -0.02 0.11 0.16 0.09 -0.06 0.06 -0.09 0.05 ...
  -0.07 -0.07 -0.10 -0.07 -0.10 -0.13 -0.09 0.03 -0.04 0.02 ...
  -0.07 -0.04 0.20 0.05 -0.02 0.12 0.06 0.04 -0.04 0.01 ...
  -0.09 -0.04 -0.14 -0.09 -0.02 -0.08 -0.09 0.05 0.05 -0.03 ...
  0.07 0.01 0.05 -0.01 0.06 -0.01 0.13 -0.04 0.03 -0.02 ...
  0.02 -0.02 0.01 -0.08 0.03 0.01 -0.10 -0.03 0.08 -0.05 ...
  0.04 -0.05 -0.07 0.09 -0.00 -0.05 0.10 -0.09 -0.01 0.01 ...

python3 softmax_classification_sgd.py --batch_size=1 --epochs=10 --learning_rate=0.005 --test_size=1597

After epoch 1: train loss 1.7683 acc 73.5%, test loss 2.0028 acc 72.2%
After epoch 2: train loss 0.7731 acc 88.5%, test loss 1.5357 acc 77.8%
After epoch 3: train loss 1.2056 acc 82.5%, test loss 2.0547 acc 73.7%
After epoch 4: train loss 1.2783 acc 88.0%, test loss 2.4955 acc 78.4%
After epoch 5: train loss 0.1834 acc 97.5%, test loss 0.8377 acc 87.8%
After epoch 6: train loss 0.5430 acc 93.0%, test loss 1.4756 acc 83.5%
After epoch 7: train loss 0.0666 acc 98.5%, test loss 0.6144 acc 89.6%
After epoch 8: train loss 0.1177 acc 95.5%, test loss 1.3177 acc 81.7%
After epoch 9: train loss 0.0033 acc 100.0%, test loss 0.6159 acc 90.9%
After epoch 10: train loss 0.0214 acc 99.0%, test loss 0.6789 acc 90.5%
Learned weights:
  -0.03 -0.10 0.04 0.14 0.09 -0.00 -0.07 0.05 0.07 -0.16 ...
  0.09 0.11 -0.24 -0.14 -0.47 0.17 -0.12 -0.06 0.02 -0.11 ...
  0.05 0.07 0.09 -0.01 -0.12 0.03 0.03 -0.10 -0.04 0.31 ...
  0.02 -0.01 0.13 0.28 0.14 0.15 -0.12 0.06 -0.08 0.18 ...
  -0.07 -0.09 -0.08 -0.33 -0.23 -0.52 -0.22 0.04 -0.04 0.04 ...
  -0.07 -0.08 0.37 0.18 0.11 0.49 0.10 0.04 -0.03 0.13 ...
  -0.09 -0.06 -0.35 -0.17 0.11 -0.02 -0.10 0.05 0.05 -0.13 ...
  0.07 0.02 0.06 0.08 0.17 0.10 0.25 -0.08 0.03 -0.08 ...
  0.02 -0.02 -0.10 -0.06 0.04 -0.12 -0.12 -0.03 0.08 -0.21 ...
  0.04 -0.04 -0.13 0.04 -0.03 -0.12 0.30 -0.08 -0.01 -0.08 ...

python3 softmax_classification_sgd.py --batch_size=100 --epochs=10 --learning_rate=0.05

After epoch 1: train loss 4.2661 acc 77.8%, test loss 4.4413 acc 75.5%
After epoch 2: train loss 0.4290 acc 90.5%, test loss 0.5414 acc 89.8%
After epoch 3: train loss 0.6189 acc 88.0%, test loss 0.5752 acc 89.2%
After epoch 4: train loss 0.3084 acc 91.9%, test loss 0.3482 acc 91.3%
After epoch 5: train loss 0.2757 acc 93.2%, test loss 0.3792 acc 91.3%
After epoch 6: train loss 0.2559 acc 92.7%, test loss 0.3718 acc 91.8%
After epoch 7: train loss 0.1164 acc 96.8%, test loss 0.1761 acc 95.1%
After epoch 8: train loss 0.2891 acc 91.5%, test loss 0.4110 acc 90.2%
After epoch 9: train loss 0.1256 acc 96.4%, test loss 0.1977 acc 94.9%
After epoch 10: train loss 0.1239 acc 96.3%, test loss 0.1847 acc 95.0%
Learned weights:
  -0.03 -0.10 -0.05 0.07 -0.08 -0.04 -0.06 0.05 0.07 -0.12 ...
  0.09 0.05 -0.24 -0.03 -0.25 0.16 -0.01 -0.06 0.02 -0.13 ...
  0.05 0.10 0.05 -0.02 -0.06 0.04 0.03 -0.10 -0.03 0.16 ...
  0.02 -0.03 0.03 0.15 0.25 0.13 -0.09 0.06 -0.09 0.11 ...
  -0.07 -0.08 -0.13 -0.10 -0.10 -0.18 -0.11 0.03 -0.04 0.00 ...
  -0.07 -0.02 0.32 0.06 0.03 0.23 0.10 0.04 -0.03 0.03 ...
  -0.09 -0.04 -0.18 -0.12 -0.01 -0.12 -0.10 0.05 0.04 -0.06 ...
  0.07 0.01 0.10 0.00 0.05 0.05 0.20 -0.02 0.03 -0.02 ...
  0.02 -0.03 -0.04 -0.12 0.02 -0.02 -0.15 -0.04 0.08 -0.08 ...
  0.04 -0.06 -0.06 0.12 -0.04 -0.10 0.12 -0.09 -0.01 0.01 ...

mlp_classification_sgd

Deadline: ~~Nov 7~~ Nov 14, 7:59 a.m. 6 points

Starting with the mlp_classification_sgd.py, implement minibatch SGD for multilayer perceptron classification.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 mlp_classification_sgd.py --epochs=3 --batch_size=10 --hidden_layer=20

After epoch 1: train acc 79.7%, test acc 80.2%
After epoch 2: train acc 91.9%, test acc 88.3%
After epoch 3: train acc 92.4%, test acc 90.0%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  -0.09 0.07 0.21 -0.16 -0.15 -0.07 0.01 -0.09 0.05 -0.11 -0.02 -0.04 ...
  -0.00 -0.00 0.00 0.00 0.00 0.00 -0.00 0.00 0.00 -0.00 0.00 -0.00 ...
  0.01 -0.01 0.01 -0.02 0.01 -0.01 0.00 0.01 0.01 -0.01 ...

python3 mlp_classification_sgd.py --epochs=3 --batch_size=10 --hidden_layer=50

After epoch 1: train acc 91.1%, test acc 89.2%
After epoch 2: train acc 95.9%, test acc 93.5%
After epoch 3: train acc 96.5%, test acc 95.2%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  0.01 0.06 -0.13 0.04 0.11 0.04 0.13 0.01 0.05 -0.05 -0.07 0.02 ...
  0.00 0.00 -0.00 0.00 -0.00 0.00 -0.00 0.00 0.00 -0.00 -0.00 -0.00 ...
  0.01 0.00 -0.00 0.00 -0.00 0.00 -0.01 -0.00 -0.00 0.00 ...

python3 mlp_classification_sgd.py --epochs=3 --batch_size=10 --hidden_layer=200

After epoch 1: train acc 95.4%, test acc 93.0%
After epoch 2: train acc 97.9%, test acc 96.6%
After epoch 3: train acc 98.8%, test acc 96.9%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  0.01 -0.09 0.04 -0.09 0.06 0.06 -0.05 -0.04 -0.00 0.02 -0.04 0.02 ...
  0.00 0.00 -0.00 -0.00 0.00 -0.00 0.00 -0.00 0.00 0.00 -0.00 -0.00 ...
  -0.00 -0.00 -0.00 -0.00 0.00 -0.00 -0.00 0.00 0.00 0.00 ...

python3 mlp_classification_sgd.py --epochs=1 --batch_size=1 --hidden_layer=200 --test_size=1597

After epoch 1: train acc 74.0%, test acc 68.7%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  0.01 -0.09 0.04 -0.09 0.06 0.06 -0.05 -0.04 -0.00 0.02 -0.04 0.02 ...
  0.00 0.00 -0.00 -0.00 -0.00 -0.00 -0.00 -0.00 0.00 -0.00 -0.00 -0.00 ...
  -0.02 0.01 -0.00 -0.02 0.02 -0.00 0.00 -0.02 0.02 0.01 ...

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 mlp_classification_sgd.py --epochs=10 --batch_size=10 --hidden_layer=20

After epoch 1: train acc 79.7%, test acc 80.2%
After epoch 2: train acc 91.9%, test acc 88.3%
After epoch 3: train acc 92.4%, test acc 90.0%
After epoch 4: train acc 96.1%, test acc 93.1%
After epoch 5: train acc 95.3%, test acc 93.1%
After epoch 6: train acc 96.6%, test acc 93.9%
After epoch 7: train acc 97.3%, test acc 94.2%
After epoch 8: train acc 98.2%, test acc 94.9%
After epoch 9: train acc 98.1%, test acc 95.7%
After epoch 10: train acc 97.4%, test acc 95.1%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  -0.07 0.12 0.33 -0.21 -0.16 -0.13 0.02 -0.14 0.01 -0.12 -0.02 -0.04 ...
  -0.00 -0.00 0.00 -0.00 0.00 0.00 -0.00 0.00 0.00 -0.00 0.00 0.00 ...
  0.02 -0.01 0.01 -0.03 0.02 -0.01 0.00 0.01 0.01 -0.01 ...

python3 mlp_classification_sgd.py --epochs=10 --batch_size=10 --hidden_layer=50

After epoch 1: train acc 91.1%, test acc 89.2%
After epoch 2: train acc 95.9%, test acc 93.5%
After epoch 3: train acc 96.5%, test acc 95.2%
After epoch 4: train acc 96.1%, test acc 94.5%
After epoch 5: train acc 96.3%, test acc 93.5%
After epoch 6: train acc 98.3%, test acc 96.2%
After epoch 7: train acc 98.4%, test acc 96.4%
After epoch 8: train acc 98.3%, test acc 95.7%
After epoch 9: train acc 99.1%, test acc 97.4%
After epoch 10: train acc 98.8%, test acc 97.4%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  0.01 0.10 -0.16 0.02 0.13 0.04 0.14 -0.01 0.05 -0.07 -0.08 0.02 ...
  0.00 0.00 -0.00 0.00 0.00 -0.00 -0.00 0.00 0.00 -0.00 -0.00 -0.00 ...
  0.01 -0.00 -0.00 0.00 0.00 0.00 -0.01 0.00 -0.00 -0.00 ...

python3 mlp_classification_sgd.py --epochs=10 --batch_size=10 --hidden_layer=200

After epoch 1: train acc 95.4%, test acc 93.0%
After epoch 2: train acc 97.9%, test acc 96.6%
After epoch 3: train acc 98.8%, test acc 96.9%
After epoch 4: train acc 98.0%, test acc 95.4%
After epoch 5: train acc 99.6%, test acc 97.7%
After epoch 6: train acc 99.7%, test acc 98.0%
After epoch 7: train acc 97.4%, test acc 95.4%
After epoch 8: train acc 99.7%, test acc 97.5%
After epoch 9: train acc 99.8%, test acc 97.9%
After epoch 10: train acc 99.9%, test acc 97.9%
Learned parameters:
  -0.03 0.09 0.05 0.02 -0.07 -0.07 -0.09 0.07 0.02 0.04 -0.10 0.09 ...
  0.01 -0.09 0.04 -0.09 0.06 0.06 -0.05 -0.04 -0.00 0.02 -0.04 0.02 ...
  0.00 0.00 0.00 -0.00 0.00 -0.00 -0.00 -0.00 0.00 0.00 0.00 -0.00 ...
  -0.00 -0.01 -0.00 -0.00 0.00 -0.00 -0.00 0.00 0.01 0.00 ...

mnist_competition

Deadline: Nov 7, 7:59 a.m. 4 points+5 bonus

This assignment is a competition task. Your goal is to perform 10-class classification on the well-known MNIST dataset. The train set contains 60k images, each consisting of $28×28$ pixels with values in $\{0, 1, …, 255\}$ . Evaluation is performed on 10k test images. You can find a simple online demo of a trained classifier here.

The mnist_competition.py template shows how to load training data, downloading it if needed. Furthermore, it shows how to save a trained estimator and how to load it during prediction.

The performance of your system is measured using accuracy of correctly predicted examples and your goal is to achieve at least 97% accuracy. Note that you can use any sklearn algorithm to solve this exercise (and of course anything you implement yourself).

multilabel_classification_sgd

Deadline: Nov 14, 7:59 a.m. 3 points

Starting with the multilabel_classification_sgd.py, implement minibatch SGD for multi-label classification and manually compute micro-averaged and macro-averaged $F_1$ -score.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 multilabel_classification_sgd.py --batch_size=10 --epochs=2 --classes=5

After epoch 1: train F1 micro 56.45% macro 46.71%, test F1 micro 58.25% macro 43.9%
After epoch 2: train F1 micro 71.46% macro 59.47%, test F1 micro 73.77% macro 60.3%
Learned weights:
  -0.05 -0.11 -0.12 -0.05 0.04 0.04 0.02 0.01 -0.05 0.03 ...
  0.05 -0.01 0.09 -0.05 -0.06 -0.08 -0.05 0.02 0.03 0.00 ...
  0.10 0.16 0.08 0.01 -0.02 -0.05 -0.11 -0.09 -0.04 0.05 ...
  0.03 0.00 -0.06 -0.01 0.01 0.06 0.10 0.08 0.12 0.01 ...
  -0.03 -0.02 -0.08 -0.05 -0.07 -0.05 0.06 -0.03 -0.09 -0.09 ...

python3 multilabel_classification_sgd.py --batch_size=10 --epochs=2 --classes=10

After epoch 1: train F1 micro 20.14% macro 9.95%, test F1 micro 21.57% macro 10.4%
After epoch 2: train F1 micro 11.29% macro 7.35%, test F1 micro 14.45% macro 8.8%
Learned weights:
  -0.04 -0.09 -0.01 -0.01 -0.09 0.02 0.01 0.04 0.02 -0.11 ...
  0.12 0.07 -0.09 -0.07 0.04 0.02 -0.06 -0.03 0.03 0.05 ...
  0.05 0.03 -0.11 -0.13 -0.09 0.08 0.02 -0.14 -0.01 -0.00 ...
  -0.03 -0.07 0.00 0.09 0.08 0.01 -0.01 -0.04 -0.08 -0.02 ...
  -0.11 -0.11 -0.04 0.04 -0.11 -0.03 -0.08 -0.03 -0.07 0.03 ...
  -0.11 -0.07 0.04 0.04 -0.00 0.04 0.00 -0.03 -0.06 -0.05 ...
  -0.14 -0.08 -0.12 -0.09 -0.11 -0.15 -0.09 -0.01 0.01 -0.05 ...
  0.04 0.00 -0.08 -0.10 -0.06 -0.04 -0.01 -0.10 -0.00 0.02 ...
  0.03 0.01 0.04 0.03 -0.06 -0.10 -0.09 0.04 0.02 -0.10 ...
  0.04 -0.06 -0.07 -0.03 -0.09 0.04 0.05 -0.09 -0.04 -0.10 ...

python3 multilabel_classification_sgd.py --batch_size=5 --epochs=2 --classes=5 --learning_rate=0.02

After epoch 1: train F1 micro 60.66% macro 47.96%, test F1 micro 60.82% macro 46.6%
After epoch 2: train F1 micro 79.28% macro 77.99%, test F1 micro 77.65% macro 71.1%
Learned weights:
  -0.08 -0.15 -0.14 -0.01 0.09 0.03 0.04 -0.08 0.03 0.08 ...
  -0.06 0.09 0.04 -0.06 -0.08 -0.13 -0.06 0.11 0.07 0.01 ...
  0.21 0.28 0.12 0.03 0.02 -0.16 -0.16 -0.14 0.06 0.13 ...
  0.07 -0.00 -0.04 0.00 0.12 0.13 0.11 0.19 0.21 0.03 ...
  0.07 -0.10 -0.10 -0.04 -0.19 0.05 0.01 -0.03 -0.15 -0.10 ...

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 multilabel_classification_sgd.py --batch_size=10 --epochs=9 --classes=5

After epoch 1: train F1 micro 56.45% macro 46.71%, test F1 micro 58.25% macro 43.9%
After epoch 2: train F1 micro 71.46% macro 59.47%, test F1 micro 73.77% macro 60.3%
After epoch 3: train F1 micro 73.06% macro 61.02%, test F1 micro 71.71% macro 56.8%
After epoch 4: train F1 micro 77.30% macro 66.48%, test F1 micro 76.19% macro 64.1%
After epoch 5: train F1 micro 76.05% macro 67.34%, test F1 micro 74.46% macro 61.4%
After epoch 6: train F1 micro 78.22% macro 73.24%, test F1 micro 77.40% macro 66.1%
After epoch 7: train F1 micro 78.13% macro 73.33%, test F1 micro 74.41% macro 61.7%
After epoch 8: train F1 micro 78.92% macro 74.73%, test F1 micro 76.78% macro 66.9%
After epoch 9: train F1 micro 80.76% macro 76.31%, test F1 micro 78.18% macro 68.3%
Learned weights:
  -0.09 -0.17 -0.16 -0.01 0.09 0.01 0.04 -0.09 0.04 0.07 ...
  -0.08 0.09 0.02 -0.07 -0.08 -0.13 -0.07 0.09 0.06 0.01 ...
  0.20 0.25 0.09 0.00 0.02 -0.18 -0.18 -0.15 0.06 0.07 ...
  0.06 -0.04 -0.07 -0.01 0.10 0.13 0.10 0.17 0.20 -0.01 ...
  0.06 -0.11 -0.12 -0.05 -0.20 0.04 -0.01 -0.03 -0.16 -0.11 ...

python3 multilabel_classification_sgd.py --batch_size=10 --epochs=9 --classes=10

After epoch 1: train F1 micro 20.14% macro 9.95%, test F1 micro 21.57% macro 10.4%
After epoch 2: train F1 micro 11.29% macro 7.35%, test F1 micro 14.45% macro 8.8%
After epoch 3: train F1 micro 41.53% macro 26.29%, test F1 micro 33.54% macro 20.4%
After epoch 4: train F1 micro 44.23% macro 30.24%, test F1 micro 37.85% macro 24.4%
After epoch 5: train F1 micro 43.23% macro 29.85%, test F1 micro 42.37% macro 28.3%
After epoch 6: train F1 micro 49.53% macro 35.63%, test F1 micro 46.53% macro 32.2%
After epoch 7: train F1 micro 55.69% macro 40.36%, test F1 micro 48.21% macro 33.8%
After epoch 8: train F1 micro 52.47% macro 37.65%, test F1 micro 46.53% macro 31.9%
After epoch 9: train F1 micro 59.89% macro 43.27%, test F1 micro 53.44% macro 37.5%
Learned weights:
  -0.02 -0.04 -0.02 -0.08 -0.04 -0.10 0.12 0.04 -0.06 -0.15 ...
  0.18 0.04 -0.10 -0.06 0.15 -0.06 -0.08 0.05 0.05 0.05 ...
  0.13 -0.02 -0.20 -0.20 -0.01 0.13 -0.06 -0.15 0.09 -0.08 ...
  -0.05 -0.08 0.11 0.12 0.13 -0.07 0.05 -0.22 -0.02 -0.02 ...
  -0.09 -0.14 -0.00 -0.02 -0.10 -0.05 -0.09 -0.08 -0.06 0.07 ...
  -0.10 -0.01 0.11 0.03 0.03 0.04 0.05 -0.11 -0.04 -0.10 ...
  -0.16 -0.09 -0.13 -0.11 -0.10 -0.20 -0.04 -0.00 0.04 -0.08 ...
  -0.03 0.05 -0.21 -0.09 -0.12 0.03 -0.13 -0.09 -0.02 0.13 ...
  0.05 0.07 0.08 0.04 -0.18 -0.11 -0.09 0.18 -0.09 -0.07 ...
  0.04 -0.10 0.00 -0.07 -0.15 0.17 -0.03 -0.12 -0.12 -0.16 ...

python3 multilabel_classification_sgd.py --batch_size=5 --epochs=9 --classes=5 --learning_rate=0.02

After epoch 1: train F1 micro 60.66% macro 47.96%, test F1 micro 60.82% macro 46.6%
After epoch 2: train F1 micro 79.28% macro 77.99%, test F1 micro 77.65% macro 71.1%
After epoch 3: train F1 micro 80.27% macro 74.86%, test F1 micro 79.57% macro 69.6%
After epoch 4: train F1 micro 81.22% macro 79.85%, test F1 micro 77.41% macro 70.1%
After epoch 5: train F1 micro 80.50% macro 78.76%, test F1 micro 72.54% macro 65.1%
After epoch 6: train F1 micro 82.86% macro 81.46%, test F1 micro 75.62% macro 69.2%
After epoch 7: train F1 micro 81.19% macro 79.54%, test F1 micro 72.51% macro 65.3%
After epoch 8: train F1 micro 81.37% macro 79.59%, test F1 micro 75.06% macro 68.9%
After epoch 9: train F1 micro 83.83% macro 82.38%, test F1 micro 79.74% macro 74.3%
Learned weights:
  -0.18 -0.31 -0.23 0.05 0.12 -0.02 0.09 -0.25 0.21 0.16 ...
  -0.21 0.18 -0.12 -0.08 -0.13 -0.17 -0.12 0.15 0.10 0.04 ...
  0.47 0.32 0.13 0.01 0.09 -0.36 -0.29 -0.26 0.27 0.14 ...
  0.12 -0.07 -0.11 0.04 0.28 0.21 0.11 0.28 0.39 0.04 ...
  0.22 -0.24 -0.26 -0.03 -0.48 0.06 -0.10 0.01 -0.28 -0.14 ...

diacritization

Deadline: Nov 14, 7:59 a.m. 5 points+5 bonus

The goal of the diacritization competition task is to learn to add diacritics to the given Czech text. We will use a small collection of fiction books, which is available under CC BY-NC-SA license. Note that these texts are the only allowed training data, you cannot use any other Czech texts (even manually annotated) to train or evaluate your model. At the test time, you will be given a text without diacritics and you should return it including diacritical marks – to be explicit, we only consider diacritized letters áčďéěíňóřšťúůýž and their uppercase variants.

The diacritization.py template shows how to load the training data, downloading it if needed.

Each sentence in the data is stored on a single line, with exactly one space character separating input words. The performance of your system is measured using word accuracy (the percentage of words you diacritized correctly, as computed by the diacritization_eval.py script) and your goal is to achieve at least 86.5%. You can use any sklearn algorithm with the exception of decision trees to solve this assignment (so no random forests, extra trees, gradient boosting, AdaBoost with decision trees, …).

tf_idf

Deadline: Nov 21, 7:59 a.m. 3 points

The template changed on Nov 8, please redownload it if you got it sooner (the LogisticRegression needs to get an additional C=10_000 argument).

Using the tf_idf.py template, perform classification of text documents from the 20 Newsgroups dataset. To represent the documents, use TF and/or IDF weights, which you implement manually (without using the sklearn.feature_extraction module in any way). Classify test set documents using sklearn.linear_model.LogisticRegression trained on the given training data, and report macro F1-score.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 tf_idf.py --train_size=1000 --test_size=500 --seed=79

Number of unique terms with at least two occurrences: 12855
F-1 score for TF=False, IDF=False: 40.7%

python3 tf_idf.py --train_size=1000 --test_size=500 --seed=79 --tf

Number of unique terms with at least two occurrences: 12855
F-1 score for TF=True, IDF=False: 44.5%

python3 tf_idf.py --train_size=1000 --test_size=500 --seed=79 --idf

Number of unique terms with at least two occurrences: 12855
F-1 score for TF=False, IDF=True: 54.1%

python3 tf_idf.py --train_size=1000 --test_size=500 --seed=79 --tf --idf

Number of unique terms with at least two occurrences: 12855
F-1 score for TF=True, IDF=True: 61.8%

python3 tf_idf.py --train_size=3000 --test_size=500 --seed=79

Number of unique terms with at least two occurrences: 26587
F-1 score for TF=False, IDF=False: 58.4%

python3 tf_idf.py --train_size=3000 --test_size=500 --seed=79 --tf

Number of unique terms with at least two occurrences: 26587
F-1 score for TF=True, IDF=False: 64.0%

python3 tf_idf.py --train_size=3000 --test_size=500 --seed=79 --idf

Number of unique terms with at least two occurrences: 26587
F-1 score for TF=False, IDF=True: 64.7%

python3 tf_idf.py --train_size=3000 --test_size=500 --seed=79 --tf --idf

Number of unique terms with at least two occurrences: 26587
F-1 score for TF=True, IDF=True: 72.8%

imdb_sentiment

Deadline: Nov 21, 7:59 a.m. 3 points

Using the imdb_sentiment.py template, perform classification of text documents from the Large Movie Review Dataset. The template also downloads pretrained word embeddings from FastText containing only words from our dataset.

Analogously to competition assignments, your task is to submit a trained model that will perform classification on the test set given to you during ReCodEx evaluation. The performance is measured using accuracy, and your goal is to achieve at least 80%.

diacritization_dictionary

Deadline: Nov 21, 7:59 a.m. 4 points+4 bonus

The diacritization_dictionary is an extension of the diacritization competition. In addition to the original training data, in this task you can also use a dictionary providing all known diacritized variants of word forms present in the training and testing data, available again under CC BY-NC-SA license. The dictionary is not guaranteed to contain all words from the training and testing data, but if it contains a word, all valid Czech diacritization variants should be present.

The rules of the competition are the same as of the diacritization competition, except that

you can utilize the dictionary, both during training and inference;
in order to pass, you need to achieve at least 95% word accuracy.

The diacritization_dictionary.py module provides a Dictionary class, which loads the dictionary (downloading it if necessary), exposing it in Dictionary.variants field as a mapping from undiacritized word form to a list of known diacritized variants.

Note that the fiction-dictionary.txt is available in ReCodEx during evaluation.

k_nearest_neighbors

Deadline: Nov 28, 7:59 a.m. 3 points

Starting with the k_nearest_neighbors.py, implement k-nearest neighbors algorithm for classifying MNIST, without using the sklearn.neighbors module or scipy.spatial module in any way.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 k_nearest_neighbors.py --k=1 --p=2 --weights=uniform --test_size=500 --train_size=100

K-nn accuracy for 1 nearest neighbors, L_2 metric, uniform weights: 73.60%

python3 k_nearest_neighbors.py --k=3 --p=2 --weights=uniform --test_size=500 --train_size=100

K-nn accuracy for 3 nearest neighbors, L_2 metric, uniform weights: 66.80%

python3 k_nearest_neighbors.py --k=1 --p=2 --weights=uniform --test_size=500 --train_size=1000

K-nn accuracy for 1 nearest neighbors, L_2 metric, uniform weights: 90.40%

python3 k_nearest_neighbors.py --k=5 --p=2 --weights=uniform --test_size=500 --train_size=1000

K-nn accuracy for 5 nearest neighbors, L_2 metric, uniform weights: 88.40%

python3 k_nearest_neighbors.py --k=5 --p=1 --weights=uniform --test_size=500 --train_size=1000

K-nn accuracy for 5 nearest neighbors, L_1 metric, uniform weights: 87.00%

python3 k_nearest_neighbors.py --k=5 --p=3 --weights=uniform --test_size=500 --train_size=1000

K-nn accuracy for 5 nearest neighbors, L_3 metric, uniform weights: 89.40%

python3 k_nearest_neighbors.py --k=1 --p=2 --weights=uniform --test_size=500 --train_size=5000

K-nn accuracy for 1 nearest neighbors, L_2 metric, uniform weights: 94.40%

python3 k_nearest_neighbors.py --k=9 --p=2 --weights=uniform --test_size=500 --train_size=5000

K-nn accuracy for 9 nearest neighbors, L_2 metric, uniform weights: 92.80%

python3 k_nearest_neighbors.py --k=9 --p=2 --weights=inverse --test_size=500 --train_size=5000

K-nn accuracy for 9 nearest neighbors, L_2 metric, inverse weights: 93.00%

python3 k_nearest_neighbors.py --k=9 --p=2 --weights=softmax --test_size=500 --train_size=5000

K-nn accuracy for 9 nearest neighbors, L_2 metric, softmax weights: 94.00%

naive_bayes

Deadline: Nov 28, 7:59 a.m. 3 points

Using the naive_bayes.py template, implement a naive Bayes classifier (without using the sklearn.naive_bayes module in any way). Support all of Gaussian NB, multinomial NB and Bernoulli NB.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 naive_bayes.py --classes=3 --naive_bayes_type=bernoulli

Test accuracy 95.17%, log probability -4933.45

python3 naive_bayes.py --classes=3 --naive_bayes_type=multinomial

Test accuracy 93.68%, log probability -300352.40

python3 naive_bayes.py --classes=3 --naive_bayes_type=gaussian

Test accuracy 95.54%, log probability -31895.82

python3 naive_bayes.py --classes=10 --naive_bayes_type=bernoulli

Test accuracy 89.21%, log probability -18342.14

python3 naive_bayes.py --classes=10 --naive_bayes_type=bernoulli --alpha=10

Test accuracy 88.54%, log probability -20829.42

python3 naive_bayes.py --classes=10 --naive_bayes_type=multinomial --alpha=10

Test accuracy 90.32%, log probability -1006524.57

python3 naive_bayes.py --classes=10 --naive_bayes_type=gaussian --alpha=10

Test accuracy 92.10%, log probability -149703.75

isnt_it_ironic

Deadline: Nov 28, 7:59 a.m. 4 points+4 bonus

The goal of the isnt_it_ironic competition task is to learn to classify given text as ironic or not.

The isnt_it_ironic.py template shows how to load the training data, downloading it if needed. Please note that the data are provided only for the purpose of this class and you cannot use them in any other way.

Each instance is a string of an English tweet. The texts have already been tokenized and tokens are separated by exactly one space. The performance of your solution will be evaluated using F1-score with sklearn.metrics.f1_score and if you surpass at least 60%, you will obtain 4 points. Note that you can use any sklearn algorithm to solve this exercise (or anything you implement yourselves).

You might find TfidfTransformer or TfidfVectorizer useful.

metric_correlation

Deadline: Dec 5, 7:59 a.m. 3 points

Using the metric_correlation.py template, find a $\beta$ for which $F_\beta$ score correlates best with human ratings.

We use an aritificial dataset, which for every sentence contains:

the number of edits that must be performed for every sentence,
the number of edits proposed by a model,
the number of correct edits proposed by a model,
human rating of the sentence.

Using bootstrap resampling, compute the mean human rating and $F_\beta$ score for each sampled dataset and then manually compute the Pearson correlation for betas between 0 and 2, and return the most correlating beta.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 metric_correlation.py --bootstrap_samples=100 --data_size=1000

Best correlation of 0.711 was found for beta 0.79

python3 metric_correlation.py --bootstrap_samples=100 --data_size=2000

Best correlation of 0.726 was found for beta 0.63

python3 metric_correlation.py --bootstrap_samples=200 --data_size=2000

Best correlation of 0.676 was found for beta 0.61

miniaturization

Deadline: Dec 5, 7:59 a.m. 2 points+4 bonus

This assignment is a competition task. Your goal is to submit the smallest model achieving at least 99% accuracy on the MNIST dataset. The total size of your submission must be at most 1 MiB, and the competition points will be awarded according to the size of your submission, which ReCodEx shows in the logs (the accuracy of your solution does not affect the competition points, as long as it is at least 99%, and is therefore not hidden as usual).

The miniaturization.py template shows how to load training data, downloading it if needed. Furthermore, it shows how to save a trained estimator and how to load it during prediction. Finally, it includes a class MLPFullDistributionClassifier, which modifies MLPClassifier to support full categorical distributions on input, i.e., each label is a distribution over the predicted classes. Such a classifier might be useful for example for knowledge distillation.

You can use any sklearn/numpy/scipy algorithm to solve this exercise (and of course anything you implement yourself).

decision_tree

Deadline: Dec 12, 7:59 a.m. 4 points

Starting with the decision_tree.py, manually implement construction of a classification decision tree, supporting both gini and entropy criteria, and max_depth, min_to_split and max_leaves constraints.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 decision_tree.py --dataset=digits --criterion=gini --min_to_split=250

Train accuracy: 60.7%
Test accuracy: 59.6%

python3 decision_tree.py --dataset=digits --criterion=gini --max_depth=3

Train accuracy: 41.1%
Test accuracy: 38.0%

python3 decision_tree.py --dataset=digits --criterion=gini --max_leaves=8

Train accuracy: 60.1%
Test accuracy: 57.1%

python3 decision_tree.py --dataset=digits --criterion=gini --min_to_split=220 --max_leaves=8

Train accuracy: 60.7%
Test accuracy: 59.6%

python3 decision_tree.py --dataset=digits --criterion=entropy --min_to_split=420

Train accuracy: 42.4%
Test accuracy: 40.2%

python3 decision_tree.py --dataset=breast_cancer --criterion=entropy --max_depth=3 --seed=44

Train accuracy: 94.8%
Test accuracy: 93.7%

python3 decision_tree.py --dataset=digits --criterion=entropy --max_leaves=7

Train accuracy: 53.2%
Test accuracy: 51.6%

python3 decision_tree.py --dataset=breast_cancer --criterion=entropy --min_to_split=55 --max_depth=3 --seed=44

Train accuracy: 94.4%
Test accuracy: 93.7%

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 decision_tree.py --dataset=iris --max_depth=3

Train accuracy: 95.5%
Test accuracy: 100.0%

python3 decision_tree.py --dataset=wine --criterion=entropy --min_to_split=20 --seed=44

Train accuracy: 97.7%
Test accuracy: 91.1%

python3 decision_tree.py --dataset=breast_cancer --criterion=entropy --max_depth=3 --seed=44

Train accuracy: 94.8%
Test accuracy: 93.7%

random_forest

Deadline: Dec 12, 7:59 a.m. 3 points

Using the random_forest.py template, train a random forest, which is a collection of decision trees trained with dataset bagging and random feature subsampling.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 random_forest.py --dataset=digits --trees=10 --max_depth=3

Train accuracy: 56.6%
Test accuracy: 55.8%

python3 random_forest.py --dataset=digits --trees=10 --bagging --max_depth=3

Train accuracy: 71.5%
Test accuracy: 71.8%

python3 random_forest.py --dataset=digits --trees=10 --feature_subsampling=0.5 --max_depth=3

Train accuracy: 69.0%
Test accuracy: 67.6%

python3 random_forest.py --dataset=digits --trees=10 --bagging --feature_subsampling=0.5 --max_depth=3

Train accuracy: 76.6%
Test accuracy: 76.9%

python3 random_forest.py --dataset=wine --trees=10 --max_depth=3

Train accuracy: 98.5%
Test accuracy: 91.1%

python3 random_forest.py --dataset=wine --trees=10 --bagging --max_depth=3

Train accuracy: 98.5%
Test accuracy: 93.3%

python3 random_forest.py --dataset=breast_cancer --trees=10 --feature_subsampling=0.5 --max_depth=3

Train accuracy: 98.6%
Test accuracy: 97.9%

python3 random_forest.py --dataset=breast_cancer --trees=10 --bagging --feature_subsampling=0.5 --max_depth=3

Train accuracy: 96.9%
Test accuracy: 96.5%

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 random_forest.py --dataset=wine --trees=3 --max_depth=3

Train accuracy: 98.5%
Test accuracy: 91.1%

python3 random_forest.py --dataset=wine --trees=3 --bagging --max_depth=3

Train accuracy: 97.0%
Test accuracy: 93.3%

python3 random_forest.py --dataset=wine --trees=3 --feature_subsampling=0.5 --max_depth=3

Train accuracy: 100.0%
Test accuracy: 97.8%

python3 random_forest.py --dataset=wine --trees=3 --bagging --feature_subsampling=0.5 --max_depth=3

Train accuracy: 98.5%
Test accuracy: 91.1%

human_activity_recognition

Deadline: Dec 12, 7:59 a.m. 2 points+4 bonus

The goal of this competition task is to perform human activity recognition, namely to recognize one of five actions (walking, standing, sitting, standing up, sitting down) using data from four accelerometers. The train set consists of 50k examples, the test set of approximately 115k.

The human_activity_recognition.py template shows how to load the training data, downloading it if needed.

Your model will be evaluated using accuracy and your goal is to achieve at least 99.25%. Note that you can use any sklearn algorithm to solve this assignment.

gradient_boosting

Deadline: Dec 19, 7:59 a.m. 6 points

Using the gradient_boosting.py template, train gradient boosted decision tree forest for classification.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 gradient_boosting.py --dataset=wine --trees=3 --max_depth=1 --learning_rate=0.3

Using 1 trees, train accuracy: 95.5%, test accuracy: 91.1%
Using 2 trees, train accuracy: 95.5%, test accuracy: 86.7%
Using 3 trees, train accuracy: 97.7%, test accuracy: 91.1%

python3 gradient_boosting.py --dataset=wine --trees=3 --max_depth=2 --learning_rate=0.3 --seed=883

Using 1 trees, train accuracy: 100.0%, test accuracy: 88.9%
Using 2 trees, train accuracy: 99.2%, test accuracy: 93.3%
Using 3 trees, train accuracy: 99.2%, test accuracy: 97.8%

python3 gradient_boosting.py --dataset=wine --trees=3 --max_depth=2 --l2=0.5 --learning_rate=0.3 --seed=488

Using 1 trees, train accuracy: 97.0%, test accuracy: 95.6%
Using 2 trees, train accuracy: 98.5%, test accuracy: 97.8%
Using 3 trees, train accuracy: 99.2%, test accuracy: 97.8%

python3 gradient_boosting.py --dataset=digits --trees=3 --max_depth=2 --learning_rate=0.5

Using 1 trees, train accuracy: 79.1%, test accuracy: 76.9%
Using 2 trees, train accuracy: 85.7%, test accuracy: 84.4%
Using 3 trees, train accuracy: 91.3%, test accuracy: 87.8%

python3 gradient_boosting.py --dataset=breast_cancer --trees=3 --max_depth=2 --learning_rate=0.5 --seed=45

Using 1 trees, train accuracy: 94.6%, test accuracy: 90.2%
Using 2 trees, train accuracy: 96.9%, test accuracy: 95.1%
Using 3 trees, train accuracy: 96.9%, test accuracy: 93.7%

nli_competition

Deadline: Dec 19, 7:59 4 points+5 bonus

In this competition task you will be solving the Native Language Identification. In that task, you get an English essay writen by a non-native individual and your goal is to identify their native language.

We will be using NLI Shared Task 2013 data, which contains documents in 11 languages. For each language, the train and test sets contain 1000 and 100 documents, respectively. Particularly interesting is the fact that humans are quite bad in this task (in a simplified settings, human professionals achieve 40-50% accuracy), while machine learning models can achieve high performance.

Because the data is not publicly available, you can download it only through ReCodEx. Please do not distribute it. The template nli_competition.py can then be used to load the dataset as usual.

The performance of your system is measured using accuracy of correctly predicted documents and your goal is to achieve at least 78% accuracy. Note that you can use any sklearn algorithm to solve this exercise.

pca

Deadline: Jan 9, 7:59 a.m. 3 points

Using the pca.py template, implement the PCA computation with both

power iteration algorithm,
SVD decomposition.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 pca.py --max_iter=5

Test set accuracy: 90.48%

python3 pca.py --max_iter=5 --pca=1

Test set accuracy: 30.28%

python3 pca.py --max_iter=5 --pca=5

Test set accuracy: 68.88%

python3 pca.py --max_iter=5 --pca=10

Test set accuracy: 80.00%

python3 pca.py --max_iter=5 --pca=20

Test set accuracy: 87.68%

python3 pca.py --max_iter=5 --pca=50

Test set accuracy: 90.28%

python3 pca.py --max_iter=5 --pca=100

Test set accuracy: 90.76%

python3 pca.py --max_iter=5 --pca=200

Test set accuracy: 90.68%

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 pca.py --max_iter=1000 --solver=lbfgs

Test set accuracy: 89.76%

python3 pca.py --max_iter=1000 --pca=1 --solver=lbfgs

Test set accuracy: 30.88%

python3 pca.py --max_iter=1000 --pca=5 --solver=lbfgs

Test set accuracy: 68.96%

python3 pca.py --max_iter=1000 --pca=10 --solver=lbfgs

Test set accuracy: 80.48%

python3 pca.py --max_iter=1000 --pca=20 --solver=lbfgs

Test set accuracy: 87.80%

python3 pca.py --max_iter=1000 --pca=50 --solver=lbfgs

Test set accuracy: 90.08%

python3 pca.py --max_iter=1000 --pca=100 --solver=lbfgs

Test set accuracy: 90.16%

python3 pca.py --max_iter=1000 --pca=200 --solver=lbfgs

Test set accuracy: 89.88%

kmeans

Deadline: Jan 9, 7:59 a.m. 3 points

Using the kmeans.py template, implement the K-Means algorithm with both

random initialization,
kmeans++ initialization.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 kmeans.py --clusters=5 --examples=150 --iterations=3 --init=random

Cluster assignments:
[4 3 4 4 3 2 3 3 4 3 4 4 4 3 1 4 3 4 1 2 3 4 3 3 1 3 3 3 3 4 0 3 3 3 4 0 3
 3 4 3 4 3 3 4 4 4 3 0 4 4 4 4 2 3 3 2 4 0 0 3 3 4 4 0 3 2 4 1 3 4 1 4 4 0
 4 3 4 1 3 0 3 4 4 3 4 1 3 4 3 3 4 4 4 3 4 4 1 4 4 3 1 3 1 4 3 3 3 4 4 4 4
 3 4 4 4 4 4 4 2 4 3 4 4 2 3 3 3 4 2 4 4 3 3 2 3 3 2 3 2 0 4 3 3 3 3 3 3 3
 3 4]

python3 kmeans.py --clusters=5 --examples=150 --iterations=3 --init=kmeans++

Cluster assignments:
[4 1 4 2 3 0 1 1 4 3 4 2 2 3 0 2 3 2 0 0 1 4 3 3 0 3 3 1 3 2 0 1 3 3 2 0 3
 3 4 1 2 3 1 2 4 2 1 0 2 4 2 2 0 1 1 0 2 0 0 1 3 4 2 0 3 0 4 0 3 2 0 4 2 0
 2 1 2 0 3 0 1 2 4 1 4 0 1 2 3 3 2 4 2 1 4 2 0 4 4 1 0 3 0 2 1 3 1 4 2 2 4
 3 2 2 4 4 2 4 0 2 1 4 4 0 3 1 3 4 0 2 4 1 1 0 3 1 0 3 0 0 4 1 1 3 3 1 1 3
 1 2]

python3 kmeans.py --clusters=7 --examples=200 --iterations=3 --init=random

Cluster assignments:
[6 0 0 3 3 1 1 1 5 4 3 6 5 2 4 4 3 3 4 3 6 3 1 3 0 6 2 0 2 6 2 1 0 3 6 3 1
 3 5 3 3 0 4 3 5 6 3 0 3 3 6 6 3 3 3 5 0 5 3 6 2 2 0 2 2 4 3 3 6 6 6 5 0 2
 1 2 2 6 2 5 0 0 1 2 5 2 2 4 5 3 6 4 6 3 3 5 1 3 3 0 5 3 0 6 2 0 2 3 5 0 3
 1 3 5 6 5 4 2 0 3 2 3 3 6 1 2 2 2 4 3 0 5 5 2 3 5 1 2 6 6 0 5 5 3 3 3 2 6
 0 0 2 0 6 3 3 2 6 0 3 5 4 1 3 5 0 3 0 5 6 3 1 5 0 0 3 0 1 3 5 4 3 3 3 2 3
 3 2 3 6 1 4 3 0 5 5 3 3 5 6 6]

python3 kmeans.py --clusters=7 --examples=200 --iterations=3 --init=kmeans++

Cluster assignments:
[4 0 0 5 5 2 2 2 2 2 5 4 2 3 2 2 6 6 2 6 4 5 2 1 0 4 3 0 3 4 3 2 0 1 4 6 2
 5 2 5 5 0 2 1 2 4 5 0 1 5 4 4 5 5 5 2 0 2 6 4 3 3 0 3 3 2 5 1 4 4 4 2 0 3
 2 3 3 4 3 2 0 0 2 3 2 3 3 2 2 5 4 2 4 5 1 2 2 5 1 0 2 1 0 4 3 0 3 1 2 0 6
 2 1 2 4 2 2 3 0 5 3 1 1 4 2 3 3 3 2 6 0 2 2 3 1 2 2 3 4 4 0 2 2 1 5 1 3 4
 0 0 3 0 4 1 5 3 4 0 1 2 2 2 6 2 0 1 0 2 4 1 2 2 0 0 5 0 2 1 2 2 1 5 1 3 1
 5 3 5 4 2 2 6 0 2 2 1 6 2 4 4]

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 kmeans.py --clusters=5 --examples=150 --iterations=5 --seed=51 --init=random

Cluster assignments:
[2 3 3 4 1 2 1 1 2 3 2 1 1 3 3 4 0 4 4 1 3 1 1 1 1 0 1 3 3 2 3 0 1 0 3 3 0
 0 1 0 1 2 1 1 3 2 1 2 2 1 3 2 2 2 3 2 1 2 1 4 3 3 4 4 2 1 1 1 1 3 1 3 1 4
 1 3 2 1 0 0 1 2 2 0 2 2 3 1 1 1 2 2 4 2 2 1 1 1 2 2 2 3 1 3 1 3 2 1 0 2 2
 3 1 1 1 3 3 0 1 3 4 1 1 4 1 3 1 4 4 3 1 4 1 4 1 1 1 3 1 1 4 2 0 3 1 4 1 2
 2 1]

python3 kmeans.py --clusters=5 --examples=150 --iterations=5 --seed=51 --init=kmeans++

Cluster assignments:
[1 3 3 4 0 1 0 2 1 3 1 2 2 3 3 4 4 4 4 2 3 2 2 2 2 4 2 3 3 0 3 4 0 4 3 3 4
 4 2 4 2 1 0 0 3 1 0 1 1 0 3 1 0 0 3 1 0 1 2 4 3 3 4 4 1 0 2 0 0 3 0 3 0 4
 2 3 1 2 4 4 2 1 1 4 1 1 3 0 2 2 1 1 4 1 1 2 0 2 1 1 1 3 0 3 2 3 1 0 4 1 1
 3 0 2 0 3 0 4 0 3 4 0 2 4 2 3 2 4 4 3 2 4 2 4 2 0 0 3 0 0 4 1 4 3 0 4 2 1
 1 2]

python3 kmeans.py --clusters=7 --examples=200 --iterations=11 --seed=67 --init=random

Cluster assignments:
[2 1 0 4 5 1 4 1 1 2 0 3 6 6 1 6 1 1 0 2 3 2 4 0 6 5 5 4 5 4 4 6 6 1 0 6 4
 4 1 6 4 5 4 4 1 0 2 1 2 2 4 3 2 1 5 2 6 0 5 6 4 2 6 3 1 1 4 5 1 2 4 5 4 5
 1 1 4 2 5 4 4 5 4 2 2 4 4 1 5 0 4 4 4 1 3 0 3 5 4 1 0 4 4 4 4 4 5 4 1 4 4
 2 5 2 6 5 2 2 4 5 4 4 3 3 2 6 1 4 1 6 1 2 3 0 5 6 4 6 4 5 5 2 0 1 6 0 1 4
 4 6 5 1 2 4 0 0 4 0 4 3 5 4 3 4 6 3 6 5 5 6 0 2 6 5 4 5 4 3 2 4 1 2 4 2 4
 2 6 4 4 6 2 4 4 6 6 5 0 2 4 1]

python3 kmeans.py --clusters=7 --examples=200 --iterations=5 --seed=67 --init=kmeans++

Cluster assignments:
[3 1 4 5 0 1 6 1 3 3 4 4 2 2 1 2 1 1 4 3 4 3 5 4 2 0 0 6 0 6 5 2 2 1 4 2 5
 5 1 2 5 0 6 6 1 4 3 1 3 3 5 4 3 1 0 3 2 4 0 2 5 3 2 4 1 1 6 0 1 3 5 0 5 0
 1 1 6 3 0 6 5 0 5 3 3 5 6 1 0 4 5 6 5 1 4 4 2 0 6 1 4 6 5 5 6 5 0 5 1 6 6
 3 0 3 2 0 3 3 5 0 6 6 4 4 3 2 1 6 1 2 1 3 4 4 0 2 6 2 6 0 0 3 4 1 2 4 1 5
 6 2 0 1 3 5 4 4 6 4 6 4 0 5 2 5 2 4 2 0 0 2 4 3 2 0 6 0 5 2 3 5 1 3 6 3 5
 3 2 6 5 2 0 6 6 2 2 0 4 3 6 1]

bootstrap_resampling

Deadline: Feb 18, 23:59 3 points

Given two trained models, compute their 95% confidence intervals using bootstrap resampling. Then, estimate the probability that the second one is better than the first one using a paired bootstrap test.

Start with the bootstrap_resampling.py template. Note that you usually need to perform a lot of the bootstrap resamplings, so you should make sure your implementation is fast enough.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 bootstrap_resampling.py --seed=47 --test_size=0.9 --bootstrap_samples=1000

Confidence intervals of the two models:
- [91.41% .. 93.88%]
- [91.96% .. 94.38%]
The estimated probability that the null hypothesis holds: 2.10%

python3 bootstrap_resampling.py --seed=47 --test_size=0.9 --bootstrap_samples=10000

Confidence intervals of the two models:
- [91.35% .. 93.88%]
- [91.97% .. 94.38%]
The estimated probability that the null hypothesis holds: 2.40%

python3 bootstrap_resampling.py --seed=47 --test_size=0.9 --bootstrap_samples=100000

Confidence intervals of the two models:
- [91.35% .. 93.88%]
- [91.97% .. 94.38%]
The estimated probability that the null hypothesis holds: 2.21%

python3 bootstrap_resampling.py --seed=55 --test_size=0.95 --bootstrap_samples=50000

Confidence intervals of the two models:
- [84.72% .. 88.00%]
- [85.30% .. 88.52%]
The estimated probability that the null hypothesis holds: 5.47%

permutation_test

Deadline: Feb 18, 23:59 2 point

Given two trained models, perform a random permutation test that the second one is better than the first one.

Start with the permutation_test.py template. Note that you usually need to perform a lot of resamplings, so you should make sure your implementation is fast enough.

Note that your results may be slightly different (because of varying floating point arithmetic on your CPU).

python3 permutation_test.py --seed=47 --test_size=0.9 --random_samples=1000

The estimated p-value of the random permutation test: 3.10%

python3 permutation_test.py --seed=47 --test_size=0.9 --random_samples=10000

The estimated p-value of the random permutation test: 2.99%

python3 permutation_test.py --seed=47 --test_size=0.9 --random_samples=100000

The estimated p-value of the random permutation test: 3.11%

python3 permutation_test.py --seed=55 --test_size=0.95 --random_samples=50000

The estimated p-value of the random permutation test: 6.55%

In the competitions, your goal is to train a model and then predict target values on the test set available only in ReCodEx.

Submitting to ReCodEx

When submitting a competition solution to ReCodEx, you should submit a trained model and a Python source capable of running it.

Furthermore, please also include the Python source and hyperparameters you used to train the submitted model. But be careful that there still must be exactly one Python source with a line starting with def main(.

Do not forget about the maximum allowed model size and time and memory limits.

Competition Evaluation

Before the deadline, ReCodEx prints the exact achieved score, but only if it is worse than the baseline.

If you surpass the baseline, the assignment is marked as solved in ReCodEx and you immediately get regular points for the assignment. However, ReCodEx does not print the reached score.
After the competition deadline, the latest submission of every user surpassing the required baseline participates in a competition. Additional bonus points are then awarded according to the ordering of the performance of the participating submissions.
After the competition results announcement, ReCodEx starts to show the exact performance for all the already submitted solutions and also for the solutions submitted later.
Each competition will be scored after the first deadline.
The bonus points will be computed in the following fashion:
- Let $B$ be the maximal number of bonus points that can be achieved in the competition;
- All of the solutions that surpass the baseline will be sorted and divided into $B+1$ groups of equal size;
- Every solution in the top group gets B points, the next group gets $B-1$ points, etc., the last group gets 0 bonus points.
- The team solution only occupies one position in the table of the competition results.
Plese, do not forget that every member of the team needs to upload the solution to ReCodEx and to submit both the training/prediction source code and the trained model itself.

What Is Allowed

You can use only the given annotated data, both for training and evaluation.
Additionally, you can use any unannotated or manually created data for training and evaluation.
The test set annotations must be the result of your system (so you cannot manually correct them; but your system can contain other parts than just trained models, like hand-written rules).
Do not use test set annotations in any way, if you somehow get access to them.
Unless stated otherwise, you can use any algorithm present in numpy or scipy, anything you implement yourself, and any pre/post-processing or ensembling methods in sklearn. Apart from the allowed algorithms, the implementation must be created by you and you must understand it fully. Do not use deep network frameworks like TensorFlow or PyTorch.

Install

Installing to central user packages repository

You can install all required packages to central user packages repository using pip3 install --user scikit-learn==1.3.0 numpy==1.24.2 scipy==1.11.2 pandas==2.1.0 matplotlib==3.7.2.
Installing to a virtual environment

Python supports virtual environments, which are directories containing independent sets of installed packages. You can create a virtual environment by running python3 -m venv VENV_DIR followed by VENV_DIR/bin/pip3 install scikit-learn==1.3.0 numpy==1.24.2 scipy==1.11.2 pandas==2.1.0 matplotlib==3.7.2 (or VENV_DIR/Scripts/pip3 on Windows).
Windows installation
- On Windows, it can happen that python3 is not in PATH, while py command is – in that case you can use py -m venv VENV_DIR, which uses the newest Python available, or for example py -3.11 -m venv VENV_DIR, which uses Python version 3.11.

Git

Is it possible to keep the solutions in a Git repository?

Definitely. Keeping the solutions in a branch of your repository, where you merge them with the course repository, is probably a good idea. However, please keep the cloned repository with your solutions private.
On GitHub, do not create a public fork with your solutions

If you keep your solutions in a GitHub repository, please do not create a clone of the repository by using the Fork button – this way, the cloned repository would be public.

Of course, if you just want to create a pull request, GitHub requires a public fork and that is fine – just do not store your solutions in it.
How to clone the course repository?

To clone the course repository, run
```
git clone https://github.com/ufal/npfl129
```
This creates the repository in the npfl129 subdirectory; if you want a different name, add it as a last parameter.

To update the repository, run git pull inside the repository directory.
How to keep the course repository as a branch in your repository?

If you want to store the course repository just in a local branch of your existing repository, you can run the following command while in it:
```
git remote add upstream https://github.com/ufal/npfl129
git fetch upstream
git checkout -t upstream/master
```
This creates a branch master; if you want a different name, add -b BRANCH_NAME to the last command.

In both cases, you can update your checkout by running git pull while in it.
How to merge the course repository with your modifications?

If you want to store your solutions in a branch merged with the course repository, you should start by
```
git remote add upstream https://github.com/ufal/npfl129
git pull upstream master
```
which creates a branch master; if you want a different name, change the last argument to master:BRANCH_NAME.

You can then commit to this branch and push it to your repository.

To merge the current course repository with your branch, run
```
git merge upstream master
```
while in your branch. Of course, it might be necessary to resolve conflicts if both you and I modified the same place in the templates.

ReCodEx

What files can be submitted to ReCodEx?

You can submit multiple files of any type to ReCodEx. There is a limit of 20 files per submission, with a total size of 20MB.
What file does ReCodEx execute and what arguments does it use?

Exactly one file with py suffix must contain a line starting with def main(. Such a file is imported by ReCodEx and the main method is executed (during the import, __name__ == "__recodex__").

The file must also export an argument parser called parser. ReCodEx uses its arguments and default values, but it overwrites some of the arguments depending on the test being executed – the template should always indicate which arguments are set by ReCodEx and which are left intact.
What are the time and memory limits?

The memory limit during evaluation is 1.5GB. The time limit varies, but it should be at least 10 seconds and at least twice the running time of my solution. For competition assignments, the time limit is 5 minutes.

Requirements

To pass the exam, you need to obtain at least 60, 75, or 90 points out of 100-point exam to receive a grade 3, 2, or 1, respectively. The exam consists of 100-point-worth questions from the list below (the questions are randomly generated, but in such a way that there is at least one question from every pair of lectures). In addition, you can get at most 40 surplus points from the practicals and at most 10 points for community work (i.e., fixing slides or reporting issues) – but only the points you already have at the time of the exam count. You can take the exam without passing the practicals first.

Exam Questions

Lecture 1 Questions

Explain why we need separate train and test data? What is generalization and how the concept relates to underfitting and overfitting? [10]
Define prediction function of a linear regression model and write down $L^2$ -regularized mean squared error loss. [10]
Starting from unregularized sum of squares error of a linear regression model, show how the explicit solution can be obtained, assuming $\boldsymbol X^T \boldsymbol X$ is regular. [20]

Lecture 2 Questions

Describe standard gradient descent and compare it to stochastic (i.e., online) gradient descent and minibatch stochastic gradient descent. [10]
Write an $L^2$ -regularized minibatch SGD algorithm for training a linear regression model, including the explicit formulas (i.e,. formulas you would need to code it in numpy) of the loss function and its gradient. [10]
Does the SGD algorithm for linear regression always find the best solution on the training data? If yes, explain under what conditions it happens, if not explain why it is not guaranteed to converge. [20]
After training a model with SGD, you ended up with a low training error and a high test error. Using the learning curves, explain what might have happened and what steps you might take to prevent this from happening. [10]
You were provided with a fixed training set and a fixed test set and you are supposed to report model performance on that test set. You need to decide what hyperparameters to use. How will you proceed and why? [5]
What method can be used for normalizing feature values? Explain why it is useful. [5]

Lecture 3 Questions

Define binary classification, write down the perceptron algorithm and show how a prediction is made for a given example. [10]
For discrete random variables, define entropy, cross-entropy, Kullback-Leibler divergence, and prove the Gibbs inequality (i.e., that KL divergence is non-negative). [20]
Explain the notion of likelihood in maximum likelihood estimation. [5]
Describe maximum likelihood estimation as minimizing NLL, cross-entropy, and KL divergence and explain the relation between them. [20]
Considering binary logistic regression model, write down its parameters (including their size) and explain how prediction is performed (including the formula for the sigmoid function). [10]
Write down an $L^2$ -regularized minibatch SGD algorithm for training a binary logistic regression model, including the explicit formulas (i.e., formulas you would to code it in numpy) of the loss function and its gradient. [20]

Lecture 4 Questions

Define mean squared error and show how it can be derived using MLE. [10]
Considering $K$ -class logistic regression model, write down its parameters (including their size) and explain how prediction is performed (including the formula for the softmax function). [10]
Explain the relationship between the sigmoid function and softmax. [5]
Write down an $L^2$ -regularized minibatch SGD algorithm for training a $K$ -class logistic regression model, including the explicit formulas (i.e., formulas you would to code it in numpy) of the loss function and its gradient. [20]
Prove that decision regions of a multiclass logistic regression are convex. [10]
Considering a single-layer MLP with $D$ input neurons, $H$ hidden neurons, $K$ output neurons, hidden activation $f$ , and output activation $a$ , list its parameters (including their size) and write down how the output is computed. [10]
List the definitions of frequently used MLP output layer activations (the ones producing parameters of a Bernoulli distribution and a categorical distribution). Then write down three commonly used hidden layer activations (sigmoid, tanh, ReLU). [10]

Lecture 5 Questions

Considering a single-layer MLP with $D$ input neurons, a ReLU hidden layer with $H$ units and a softmax output layer with $K$ units, write down the explicit formulas (i.e., formulas you would to code it in numpy) of the gradient of all the MLP parameters (two weight matrices and two bias vectors), assuming input $\boldsymbol x$ , target $t$ , and negative log likelihood loss. [20]
Formulate the Universal approximation theorem. [10]
How do we search for a minimum of a function $f(\boldsymbol x): \mathbb{R}^D \rightarrow \mathbb{R}$ subject to equality constraints $g_1(\boldsymbol x)=0, \ldots, g_m(\boldsymbol x)=0$ ? [10]
Prove which categorical distribution with $N$ classes has maximum entropy. [10]
Consider derivation of softmax using maximum entropy principle, assuming we have a dataset of $N$ examples $(x_i, t_i), x_i \in \mathbb{R}^D, t_i \in \{1, 2, \ldots, K\}$ . Formulate the three conditions we impose on the searched $\pi: \mathbb{R}^D \rightarrow \mathbb{R}^K$ , and write down the Lagrangian to be minimized. [20]
Define precision (including true positives and others), recall, $F_1$ score, and $F_\beta$ score (we stated several formulations for $F_1$ and $F_\beta$ scores; any one of them will do). [10]
Explain the difference between micro-averaged and macro-averaged $F_1$ scores. [10]
Explain (using examples) why accuracy is not a suitable metric for unbalanced target classes, e.g., for a diagnostic test for a contagious disease. [5]

Lecture 6 Questions

Explain how is the TF-IDF weight of a given document-term pair computed. [5]
Define conditional entropy, mutual information, write down the relation between them, and finally prove that mutual information is zero if and only if the two random variables are independent (you do not need to prove statements about $D_\textrm{KL}$ ). [10]
Show that TF-IDF terms can be considered portions of suitable mutual information. [10]
Explain the concept of word embedding in the context of MLP and how it relates to representation learning. [5]
Describe the skip-gram model trained using negative sampling. [10]
How would you proceed to train a part-of-speech tagger (i.e., you want to assign each word with its part of speech) if you only could use pre-trained word embeddings and MLP classifier? [5]

Lecture 7 Questions

Describe $k$ -nearest neighbors prediction, both for regression and classification. Define $L_p$ norm and describe uniform, inverse, and softmax weighting. [10]
Show that $L^2$ -regularization can be obtained from a suitable prior by Bayesian inference (from the MAP estimate). [10]
Write down how $p(C_k | \boldsymbol x)$ is approximated in a Naive Bayes classifier, explicitly state the Naive Bayes assumption, and show how is the prediction performed. [10]
Considering a Gaussian naive Bayes, describe how are $p(x_d | C_k)$ modeled (what distribution and which parameters does it have) and how we estimate it during fitting. [10]
Considering a Bernoulli naive Bayes, describe how are $p(x_d | C_k)$ modeled (what distribution and which parameters does it have) and how we estimate it during fitting. [10]

Lecture 8 Questions

Prove that independent discrete random variables are uncorrelated. [10]
Write down the definition of covariance and Pearson correlation coefficient $\rho$ , including its range. [10]
Explain how are the Spearman's rank correlation coefficient and the Kendall rank correlation coefficient computed (no need to describe the Pearson correlation coefficient). [10]
Describe setups where a correlation coefficient might be a good evaluation metric. [5]
Describe under what circumstance correlation can be used to assess validity of evaluation metrics. [5]
Define Cohen's $\kappa$ and explain what it is used for when preparing data for machine learning. [10]
Considering an averaging ensemble of $M$ models, prove the relation between the average mean squared error of the ensemble and the average error of the individual models, assuming the model errors have zero means and are uncorrelated. [20]
Explain knowledge distillation: what it is used for, describe how it is done. [10]

Lecture 9 Questions

In a regression decision tree, state what values are kept in internal nodes, define the squared error criterion and describe how is a leaf split during training (without discussing splitting constraints). [10]
In a $K$ -class classification decision tree, state what values are kept in internal nodes, define the Gini index and describe how is a node split during training (without discussing splitting constraints). [10]
In a $K$ -class classification decision tree, state what values are kept in internal nodes, define the entropy criterion and describe how is a node split during training (without discussing splitting constraints). [10]
For binary classification, derive the Gini index from a squared error loss. [20]
For $K$ -class classification, derive the entropy criterion from a non-averaged NLL loss. [20]
Describe how is a random forest trained (including bagging and a random subset of features) and how is prediction performed for regression and classification. [10]

Lecture 10 Questions

Write down the loss function which we optimize in gradient boosted decision trees during the construction of $t^\mathrm{th}$ tree. Then define $g_i$ and $h_i$ and show the value $w_\mathcal{T}$ of optimal prediction in node $\mathcal{T}$ and the criterion used during node splitting. [20]
For a $K$ -class classification, describe how to perform prediction with a gradient boosted decision tree trained for $T$ time steps (how the individual trees perform prediction and how are the $K \cdot T$ trees combined to produce the predicted categorical distribution). [10]
What type of data are gradient boosted decision trees good for as opposed to multilayer perceptron? Explain the intuition why it is the case. [5]

Lecture 11 Questions

Formulate SVD decomposition of matrix $\boldsymbol X$ , describe properties of individual parts of the decomposition. Explain what the reduced version of SVD is. [10]
Formulate the Eckart-Young theorem. [10]
Explain how to compute the PCA of dimension $M$ using the SVD decomposition of a data matrix $\boldsymbol X$ , and why it works. [10]
Given a data matrix $\boldsymbol X$ , write down the algorithm for computing the PCA of dimension $M$ using the power iteration algorithm. [20]
Describe the K-means algorithm, including the kmeans++ initialization. [20]

Lecture 12 Questions

Considering statistical hypothesis testing, define type I errors and type II errors (in terms of the null hypothesis). Finally, define what a significance level is. [10]
Explain what a test statistic and a p-value are. [10]
Write down the steps of a statistical hypothesis test, including a definition of a p-value. [10]
Explain the differences between a one-sample test, two-sample test, and a paired test. [10]
When considering multiple comparison problem, define the family-wise error rate, and prove the Bonferroni correction, which allows limiting the family-wise error rate by a given $\alpha$ . [10]
For a trained model and a given test set with $N$ examples and metric $E$ , write how to estimate 95% confidence intervals using bootstrap resampling. [10]
For two trained models and a given test set with $N$ examples and metric $E$ , explain how to perform a paired bootstrap test that the first model is better than the other. [10]
For two trained models and a given test set with $N$ examples and metric $E$ , explain how to perform a random permutation test that the first model is better than the other with a significance level $\alpha$ . [10]

Lecture 13 Questions

Explain the difference between deontological and utilitarian ethics. List examples on how these theoretical frameworks can be applied in machine learning ethics. [10]
List a few examples of potential ethical problems related to data collection. [5]
List a few examples of potential ethical problems that can originate in model evaluation. [5]

Related Courses

Deep Learning by Milan Straka

Course introducing deep neural networks, from the basics to the latest advances, focusing both on theory as well as on practical aspects.

Deep Reinforcement Learning by Milan Straka

Course introducing reinforcement learning, from basic tabular methods to involvement of deep neural networks, focusing both on theory as well as on practical aspects.

Unsupervised Machine Learning in NLP by David Mareček

The seminar focuses on deeper understanding of selected unsupervised machine learning methods for students who already have basic knowledge of machine learning and probability models.

Search form

Introduction to Machine Learning with Python – Winter 2023/24

About

Timespace Coordinates

Course Objectives

Lectures

License

1. Introduction to Machine Learning

2. Linear Regression, SGD

3. Peceptron, Logistic Regression

4. Multiclass Logistic Regression, Multilayer Perceptron

5. MLP, Softmax as MaxEnt classifier, F1 score

6. Representing Text (TF-IDF, Word2Vec)

7. K Nearest Neighbors, Naive Bayes

8. Correlation, Model Combination

9. Decision Trees, Random Forests

10. Gradient Boosted Decision Trees

11. SVD, PCA, k-means

12. Statistical Hypothesis Testing, Model Comparison

13. Machine Learning Ethics, Final Summary

Requirements

Environment

Teamwork

No Cheating

linear_regression_manual

linear_regression_features

linear_regression_l2

linear_regression_sgd

feature_engineering

rental_competition

perceptron

logistic_regression_sgd

grid_search

thyroid_competition

softmax_classification_sgd

mlp_classification_sgd

mnist_competition

multilabel_classification_sgd

diacritization

tf_idf

imdb_sentiment

diacritization_dictionary

k_nearest_neighbors

naive_bayes

isnt_it_ironic

metric_correlation

miniaturization

decision_tree

random_forest

human_activity_recognition

gradient_boosting

nli_competition

pca

kmeans

bootstrap_resampling

permutation_test

Submitting to ReCodEx

Competition Evaluation

What Is Allowed

Install

Git

ReCodEx

Requirements

Exam Questions

Related Courses

Deep Learning by Milan Straka

Deep Reinforcement Learning by Milan Straka

Unsupervised Machine Learning in NLP by David Mareček

Archive

Winter 2022/23

Winter 2021/22

Winter 2020/21

Winter 2019/20