Scripting library

[RokIvansek]

Implement a library interactions.py

• [x] Within a library, implement a class Interactions. Initialization should accept an Orange data table (warning: the table could be sparse, check text mining add-on and speak with Niko Colnerič) and an optional discretizer (default is equal-interval discretization with three intervals). Initialization should discretize the data and compute the information gain for all attributes (get_information_gain). Information gains are stored in a dictionary self.info_gains.
• [x] A method attribute_interaction(a, b) [a and b are attribute names, or indices, or anything that Orange can use to access the attributes] returns the object within interaction information (abs_interaction, rel_interaction, ig_a, ig_b, ig_ab, …). That is, all values that are needed to compute interaction chart. Make sure that the returned objects prints nicely (str() or print()).
• [x] A method interaction_matrix returns a matrix of all interactions. Use the same type of object that Orange uses for distance matrix. Using widgets, this object can be visualized in Distance Map or Distance Matrix.

[RokIvansek]

Isn't equal-frequencies interval discretization a better choice? Which of the interaction information (abs_interaction, rel_interaction, ig_a, ig_b, ig_ab, …) should be stored in the interaction matrix?

[RokIvansek]

Equal-frequencies was used. Value rel_ig_ab is stored in an interaction matrix.