mksroczynska / introtomachinelearning

My solutions of the tasks from the book: Miroslav Kubat "An Introduction to Machine Learning".

Intro To Machine Learning

My solutions of the tasks from the book: Miroslav Kubat An Introduction to Machine Learning.

1.1 Hill climbing for sliding-tile.

Starting from some random initial state of a sliding-tile trying to obtain a given final state using hill climbing algorithm. The evaluation function is set to be the sum of distances of every number from its position in the final state calculated as |x - x_final| + |y-y_final| for a single point.

2.2 Using Bayes formula to calculate the class probabilities in a domain where all attributes are discrete.

For an object decribed by the given set of attributes, the program says what is its most probable class. The test values is data from pies.csv and an object described by the attributes from this domain.

Data: pies.csv

The dataset for pies domain described in chapter 1 of the book.

2.3 Bayes for continuous attributes.

Data: example1.csv

A set consists of six examples, each described by three continuous attributes: at1, at2, at3, class.

Data: wines.csv

The dataset from http://archive.ics.uci.edu/ml/machine-learning-databases/wine/.

3 Nearest neighbours

3.1 and 3.2

Simple classification using nearest neighbours method. Test are performed on the normalized data from wines.csv. The dependence between the efficiency and number of nearest neighbours taken into account is plotted.

3.3

Another test is done on a synthetic domain of points, where each is described by two continuous attributes, x and y in range [0,1]. The points are classified as pos if they are inside a circle of some arbitrary radius, and neg otherwise. For some amount of randomly chosen points, the names are swapped and thus, the noise is introduced. The plots show how well the k-NN method works for such cases.

3.4

It is useful to try to minimize the noise. This is done with Tomek-links method.

4. Linear and polynomial classifiers.

To be done

5. Artificial neural networks.

To be done

6. Decision trees as classifiers

6.1 Induction of decision trees.

6.2 Pruning mechanism.

6.3 Computational time of induction.