edunuke

def init(params):
    try:
        regularizer = params["regularizer"]
        constraint = keras.constraints.UnitNorm(axis=0)
     
        risk_embed_dim =int(np.round(len(set(risk_df.credit_score)) ** (1/4)))
        products_embed_dim =int(np.round(len(set(products_df.product_code)) ** (1/4)))
        user_embed_dim =int(np.round(len(set(users_df.client_id)) ** (1/4)))
        # user branch
        user_input = keras.layers.Input(shape=(user_shape[-1]-1,), name='user_input')

edunuke

# Hyperparameter search space
params= {
        'nodes1': hp.choice('n1',[100,]),
        'nodes2': hp.choice('n2',[200]),
        'nodes3': hp.choice('n3',[300]),
        'nodes4': hp.choice('n4', [200]),
        'nodes5': hp.choice('n5', [100]),
        'batch_size': hp.choice('bs', [1000]),
        'regularizer': hp.choice('reg', [None,]),
        'optimizer': hp.choice('opt', [keras.optimizers.Adam(lr=0.001,

edunuke

# set up the input data so it can be read by the neural network
input_train = [user_train.drop(['client_id'],axis=1).values,
               product_train.drop(['client_id'],axis=1).values,
               risk_train.drop(['client_id'],axis=1).values]

input_val = [user_val.drop('client_id',axis=1).values,
             product_val.drop(['client_id'],axis=1).values,
             risk_val.drop('client_id',axis=1).values]

edunuke

# split dataset 80/20
traintest_split = model_selection.train_test_split(products_df.drop(TARGET_NAME,axis=1),
                                                  risk_df,
                                                  users_df,
                                                  y_class, test_size=0.2, random_state=SEED)

# Unpack the result in their individual parts
product_train, product_val, risk_train, risk_val, user_train, user_val, y_class_train, y_class_val, = traintest_split

edunuke

#target variable vector
y_class = pd.get_dummies(products_df[TARGET_NAME])

# print(y_class[0].values)
#[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

edunuke

# load datasets
users_df = pd.read_csv("dataset/users_data.csv", index_col=0)
risk_df = pd.read_csv("dataset/risk_data.csv", index_col=0)
products_df = pd.read_csv("dataset/products_data.csv", index_col=0)


# Static Constants
SEED = 123
TARGET_NAME = "product_code"
RESAMPLED= False

edunuke

import os
import numpy as np
import pandas as pd
import pickle
import matplotlib.pyplot as plt
import keras
import keras.backend as K
from sklearn.utils import class_weight
from sklearn import dummy, metrics, model_selection, datasets, preprocessing
from sklearn.manifold import TSNE

edunuke

from lime import submodular_pick

#set up sp lime with 20 samples. The more amount of samples time increases dramatically
sp_obj = submodular_pick.SubmodularPick(explainer, 
                                        X_train.values[:500],
                                        predict, 
                                        sample_size=20,
                                        num_features=9,
                                        num_exps_desired=5)

edunuke

import lime
import lime.lime_tabular
from sklearn import linear_model

# set up the LIME explainer
explainer = lime.lime_tabular.LimeTabularExplainer(X_train,
                                                  training_labels = Y_train.values,
                                                  feature_names = X_train.columns,
                                                  mode = 'regression',
                                                  discretize_continuous = False)

edunuke

# get the ovearall mean contribution of each feature variable
shap.summary_plot(shap_values[0], X_train.values[:500], feature_names=X_train.columns)