Machine Learning

Exercise Content

• Generate all combinations without repetition of length 3 from 6 possible ones.
• On one of the test files (chess, connect, ...), generate frequent patterns and calculate Support.
• From the generated frequent patterns, write down the rules and their Confidence.

Exercise Content

• Download a dataset from Kaggle or UCI Machine Learning Repository.
• Explore the dataset and understand them with selected plots.
• Apply data transformation techniques to prepare the data for analysis.

Exercise Content

• Implement two agglomerative clustering approaches – single linkage and complete linkage – using Manhattan and Euclidean metrics derived from a distance matrix.
• Stop criteria: either terminate when a predefined number of clusters is reached or carry out full clustering and then cut the dendrogram at the appropriate level to yield the desired number of clusters.
• Visualization: present the resulting cluster assignments using a scatter plot to display the clustering structure.

Exercise Content

• Download a dataset from Kaggle or UCI Machine Learning Repository.
• Apply K-means, Hierarchical and DB scan algorithms.
• Compare the results and interpret the results.

Exercise Content

• Check the datasets in this page for experimenting.
• Apply SVD (complete) and Reduced SVD method to Bars Dataset and evaluate also data with noise.
• Reconstruct the data with different reduction (2, 5, 10, 16 basis functions).
• Try the NNMF method similarly.
• Take the Mnist dataset and apply SVD recomposition and visualize it in vector space. Coloring take from Labels
• Try to reconstruct the reduced data and visualize it.
• Use PCA and t-SNE for further visualization of the Mnist dataset.

Exercise Content

• Implement a decision tree from scratch.
• Use a simple dataset to train the decision tree.
• Visualize the decision tree structure (console output is enough).
• Evaluate the performance of the decision tree.

Exercise Content

• Apply the classification algorithms from Scikit-learn.
• Use classical datasets to train different models.
• Preprocessing the data to gain maximum performance.
• Evaluate the performance of the model.

Exercise Content

• Apply the regression algorithms from Scikit-learn.
• Use classical dataset or presented dataset to train different models.
• Preprocessing the data to gain maximum performance.
• Evaluate the performance of the model.

Exercise Content

• Apply the resampling techniques from Scikit-learn.
• Use classical dataset or presented dataset to train different models.
• Preprocessing the data to gain maximum performance.
• Evaluate the performance of the models.

Exercise Content

• Load a proper dataset.
• Prepare data for forecasting tasks.
• Preprocessing the data to gain maximum performance.
• Evaluate the performance of the models.

Exercise Content

• The first part of this exercise requires you to develop a machine learning model for Titanic survival prediction by loading the Titanic dataset, performing necessary data preprocessing (without being obligated to use all records or attributes), and then training any classification model of your choosing.
• For robustness, it is suggested to use tools from sklearn.pipeline or sklearn.compose to integrate the preprocessing steps and the model into a unified object.
• Once the model is trained, it needs to be saved, which must then be serialized—for instance, using Pickle—to allow the trained model to be easily loaded and utilized later in the web application.
• Create a web application/service for obtaining predictions. - One option is to use Flask to create a simple API that returns the model's prediction for given parameters in JSON format.
• Test your web application with predictions for real or invented inputs (e.g., 2 real examples from the dataset and 1 invented):
• {“Age”: 22, “Fare”: 7.25, “SibSp”: 1, “Parch”: 0, “Pclass”: 3, “Embarked”: “S”, “Sex”: “male”}
• {“Age”: 38, “Fare”: 71.2833, “SibSp”: 1, “Parch”: 0, “Pclass”: 1, “Embarked”: “C”, “Sex”: “female”}
• {“Age”: 18, “Fare”: 10, “SibSp”: 1, “Parch”: 2, “Pclass”: 2, “Embarked”: “S”, “Sex”: “male”}

Available datasets

• Test
• Chess
• Connect
• Mushroom
• T10I4D100K (generated shopping cart)

Available datasets

• Clusters-3
• Clusters-5
• Clusters-5 overlapping
• Annulus
• Boxes
• Dense Grid

Available datasets

The dataset Bars contains images of bars with 8x8 pixels with different levels of noise. Mnist is in the shape of 28x28 pixels, with lables stored separatelly in Mnist labels dataset.

• Bars
• Bars 20% noise
• Bars 50% noise
• Mnist handwritten digits
• Mnist Labels

Available datasets

• Separable
• Non separable
• IRIS dataset
• Annulus
• Boxes
• 5-clusters
• Dense grid
• Tic-Tac-Toe

Available datasets

• Bike
• Forrest Fires
• Wine Quality - red
• Wine Quality - white