sa757 / handset-model-product-nn

Telenor Handset Model with Product-based Neural Networks

handset-model-product-nn

Telenor Handset Model with Product-based Neural Networks

This repository explores a binary classification problem through...

1. feature selection
2. xgboost
3. wide&deep (Google)
4. complementary neural networks (cmtnn)
5. product-based neural networks (pnn) (main focus)

note 1: the relevant papers are included in the papers directory

note 2: the company data has not been included in this repository

For an overview, refer to the presentation directory.

The specific results obtained from different models are in the results directory.

---

label column: "TARGET_S_TO_S_APPLE"

Split IDs and their contents (produced via handset_model_current.py):

• id=1: all features
  • dropped: ['Unnamed: 0', 'ID', 'MPP_NET_DISCOUNT_OTHER_FEE']
  • categorical and binary encoded as -1/1
  • standardized numerical

standardized numerical and categorical/binary encoded as 0/1:

• id=2: sklearn selectKBest w/ f_classif features
• id=3: all cat features
• id=4: all features selected in feature_selection directory
• id=5: f_classif features, cat vars without any encoding
• id=6: features from COLS_TERM (look in handset_model_current.py)

note: refer to features in the feature_selection directory

---

note: handset_model_original.py was written for python 2.7 while handset_model_current.py has been debugged (and further modified) for python 3.6