permImp.py

# import libraries
import numpy as np
import pandas as pd
import eli5

from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
from eli5.sklearn import PermutationImportance

# load data file
train = pd.read_csv('50-Startups.csv')

# perform one-hot encoding for categorical variable
trainDummies = pd.get_dummies(train['State'], prefix = 'state')

# combine original and one-hot encoded dataframes together
train = pd.concat([train, trainDummies], axis=1)

# extract dependent (predictor) feature
y = train.Profit

# extract independent features from combined dataframe by removing unwanted features
X = train.drop(["Profit", "State", "state_New York"], axis=1)

# split train and validation datasets
train_X, val_X, train_y, val_y = train_test_split(X, y, random_state=5)

# create Random Forest Regressor model and fit dependent and independent features
rf_model = RandomForestRegressor(random_state=0).fit(train_X, train_y)

# compute Permutation Importance using Random Forest Regressor model on validation dataset 
permImportance = PermutationImportance(rf_model, random_state=0).fit(val_X, val_y)

# print computer feature weights
eli5.show_weights(permImportance, feature_names = X.columns.tolist())