README.md

In signaux-faibles/fte: Feature Target Encoding for classification

fte

Feature Target Encoding for classification.

Currently only support two class classification and simple target encoding based on empirical means.

Current developments: * Two-level hierarchical bayesian model encoding for hierarchical categories.

How to use it

• First create a data preparation map with function target_encode_create, with your train data only.

• Then apply your preparation map to your different samples:

• For your train data, to avoid overfitting for small samples, you can add some noise with noise_level, ignore the data of the entry the target encoding is applied to with holdout_type = "leave_one_out", press the target encoding towards the whole sample mean using prior_sample_size (useful again for small categories).
• For your test_data, always put holdout_type = "none" and noise_level = 0 (default values)

signaux-faibles/fte documentation built on Jan. 29, 2020, 8:07 p.m.