feature_importance.py

import joblib
import pickle
import sklearn.inspection
import numpy as np
import pandas as pd
import os.path


rows = ["Pers Clubs A","Pers Clubs 10","Pers Clubs K","Pers Clubs Q","Pers Clubs J",
        "Pers Diamonds A","Pers Diamonds 10","Pers Diamonds K","Pers Diamonds Q","Pers Diamonds J",
        "Pers Hearts A","Pers Hearts 10","Pers Hearts K","Pers Hearts Q","Pers Hearts J",
        "Pers Spades A","Pers Spades 10","Pers Spades K","Pers Spades Q","Pers Spades J",
        "P1 points","P2 points","P1 pending points","P2 pending points","Trump suit","Phase",
        "Stock size","Leader","Whose turn","Opponents played card"]

feature_df = pd.DataFrame(rows,columns=["Feature"])
df_list = []

model_names = ["BNB","LR","MLPC"]
model_dataset = "_rand_g50000_f1"

for model_name in model_names:
    #feature permutation importance file name:
    fpi_file = "./fpi/fpi%s_%s.csv" %(model_dataset, model_name)

    if not os.path.isfile(fpi_file):
        evaluate_dataset = "_rand_g10000_f1"
        with open("./datasets/dset%s.pkl" %evaluate_dataset, 'rb') as output:
            data, target = pickle.load(output)

        model = joblib.load("./models/m%s_%s.pkl" %(model_dataset, model_name))
        result = sklearn.inspection.permutation_importance(model, data, target)
        np.savetxt("./fpi/fpi%s_%s.csv" %(model_dataset, model_name), result.importances_mean, delimiter=",")

    index = 0
    feature_importance = np.array([])
    data = np.genfromtxt(fpi_file, delimiter=',')

    #Perspective for 20 cards, each can be in of 6 states: Unknown, in Stock, P1 Hand, P2 Hand, P1 Won, P2 Won
    for i in range(20):
        feature_importance = np.append(feature_importance, np.average( data[index:index+6] ))
        index += 6

    # P1_points,P2_points,P1_pending_points,P2_pending_points
    for i in range(4):
        feature_importance = np.append(feature_importance, data[index])
        index += 1

    # Trump_suit
    feature_importance = np.append(feature_importance, np.average( data[index:index+4] ))
    index += 4

    # Phase
    feature_importance = np.append(feature_importance, np.average( data[index:index+2] ))
    index += 2

    # Stock_size
    feature_importance = np.append(feature_importance, data[index])
    index += 1

    # Leader
    feature_importance = np.append(feature_importance, np.average( data[index:index+2] ))
    index += 2

    # Whose_turn
    feature_importance = np.append(feature_importance, np.average( data[index:index+2] ))
    index += 2

    # 21 Opponents_played_cards (21 because opponent can play 20 cards or No card if we start the trick)
    feature_importance = np.append(feature_importance, np.average( data[index:index+21] ))
    index += 21

    # Normalize and round
    feature_importance = np.round(feature_importance/np.sum(feature_importance),4)

    df = pd.concat( [feature_df, pd.DataFrame(feature_importance,columns=[model_name]) ], axis=1)
    df_sorted = (df.sort_values(model_name, ascending=False)).reset_index(drop=True)
    df_list.append(df_sorted)


final_df = df_list[0]
for i in range(1,len(df_list)):
    final_df = pd.concat([final_df, df_list[i]], axis=1)

print(final_df)