-
Notifications
You must be signed in to change notification settings - Fork 0
/
rfr_housing.py
220 lines (164 loc) · 8.54 KB
/
rfr_housing.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
import pandas as pd
import numpy as np
import sklearn.model_selection as skms
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import make_scorer, r2_score
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
def rmse(y_true, y_pred, data_df, weight_var_name=''):
# returns root mean squared error
# if weight=0: rmse = sqrt(sum((y_true-y_pred)^2))
# if weighted: rmse = sqrt(sum(weight*(y_true-y_pred)^2)/sum(weight))
if weight_var_name == '':
df = pd.DataFrame({'y_true': y_true, 'y_pred': y_pred})
df['weight'] = 1
else:
weight = data_df[weight_var_name].loc[y_true.index.values]
df = pd.DataFrame({'weight': weight, 'y_true': y_true, 'y_pred': y_pred})
rmse_ = np.sqrt(np.sum(df['weight']*(df['y_true']-df['y_pred'])**2)/np.sum(df['weight']))
return rmse_
def rmae(y_true, y_pred, data_df, weight_var_name=''):
# returns root mean absolute error
# if weight=0: rmae = sqrt(sum(abs(y_true-y_pred)))
# if weighted: rmse = sqrt(sum(weight*(abs(y_true-y_pred)))/sum(weight))
if weight_var_name == '':
df = pd.DataFrame({'y_true': y_true, 'y_pred': y_pred})
df['weight'] = 1
else:
weight = data_df[weight_var_name].loc[y_true.index.values]
df = pd.DataFrame({'weight': weight, 'y_true': y_true, 'y_pred': y_pred})
#if weight==0:
# rmae = (np.sum(weight*np.absolute(y_true.values-y_pred)))**0.5
#else:
# rmae_ = (np.sum(weight*np.absolute(y_true.values-y_pred))/np.sum(weight))
rmae_ = np.sqrt(np.sum(df['weight']*np.absolute(df['y_true']-df['y_pred']))/np.sum(df['weight']))
return rmae_
def get_rmse_pctl(y_true, y_pred, weight, var, var_dict):
# weighted rmse by percentiles of a variable of interest
# -- returns dict {pct: rmse}
weight = weight.loc[y_true.index.values]
var = var.loc[y_true.index.values]
df = pd.DataFrame({'var': var, 'weight': weight, 'y_true': y_true, 'y_pred': y_pred})
dict_ = {}
for pct, threshold in var_dict.items():
if pct<100:
df_temp = df[df['var']<threshold]
pct_rmse = np.sqrt(((df_temp['weight']*(df_temp['y_true']-df_temp['y_pred'])**2).sum())/df_temp['weight'].sum())
dict_[pct] = pct_rmse
return dict_
if __name__ == '__main__':
seed = 5941
#---- features and target variable
quant_features = ['bedrooms', 'bathrooms', 'sqft_living', 'sqft_lot', 'grade', 'age', 'appliance_age', 'crime', 'renovated']
cat_features = ['backyard', 'view', 'condition']
features = quant_features
target_var = 'price'
weight = 'weight'
sqft = 'sqft_living'
#---- uploading data
heart_df = pd.read_csv('data\\housing.csv')
rows, cols = heart_df.shape
print(f'> rows = {rows}, cols = {cols}')
#---- get weighted percentiles
#sqft_pctl_dict = wgt_percentile(heart_df, weight, sqft)
#target_pctl_dict = wgt_percentile(heart_df, weight, target_var)
#---- train/test split
X, y = heart_df[features], heart_df[target_var]
X_train, X_test, y_train, y_test = skms.train_test_split(X, y, test_size=0.20, random_state = seed)
X_train_rows, y_train_rows = X_train.shape[0], y_train.shape[0]
X_test_rows, y_test_rows = X_test.shape[0], y_test.shape[0]
train_rows, test_rows = -1, -1
if X_train_rows == y_train_rows:
train_rows = X_train_rows
if X_test_rows == y_test_rows:
test_rows = X_test_rows
X_train_weights = heart_df[weight].loc[X_train.index.values]
X_test_weights = heart_df[weight].loc[X_test.index.values]
#X_train_sqft = heart_df[sqft].loc[X_train.index.values]
#X_train_target_var = heart_df[target_var].loc[X_train.index.values]
params_score = {"data_df": heart_df}
print(f'> features = {len(features)}')
print(f'> training set = {train_rows} ({round(train_rows*1.0/rows,3)})')
print(f'> testing set = {test_rows} ({round(test_rows*1.0/rows,3)}) \n')
#---- weighted scorer functions
wgt_rmse_scorer = make_scorer(rmse, greater_is_better=False, **params_score, weight_var_name=weight)
wgt_rmae_scorer = make_scorer(rmae, greater_is_better=False, **params_score, weight_var_name=weight)
scorers = {'rmse': wgt_rmse_scorer, 'rmae': wgt_rmae_scorer}
#---- random forest training with hyperparameter tuning
pipe = Pipeline([("scaler", StandardScaler()), ("rfr", RandomForestRegressor())])
random_grid = { "rfr__n_estimators": [100, 500, 1000],
"rfr__max_depth": [10, 20, 30],
"rfr__max_features": [0.25, 0.50, 0.75],
"rfr__min_samples_split": [15, 25],
"rfr__min_samples_leaf": [5, 10, 15],
"rfr__bootstrap": [True, False]
}
optimized_rfr = skms.RandomizedSearchCV(pipe,
param_distributions=random_grid,
n_iter = 50,
cv = 5,
verbose = 10,
scoring = scorers,
refit = 'rmse',
random_state = seed,
n_jobs = 4)
optimized_rfr.fit(X_train, y_train, **{'rfr__sample_weight': X_train_weights.values.ravel()})
print('\n')
#---- predicting on the training and testing set
y_train_pred = optimized_rfr.predict(X_train)
rmse_train = round(rmse(y_train, y_train_pred, heart_df, weight), 0)
rmae_train = round(rmae(y_train, y_train_pred, heart_df, weight), 0)
r2_train = round(r2_score(y_train, y_train_pred, X_train_weights), 4)
y_test_pred = optimized_rfr.predict(X_test)
rmse_test = round(rmse(y_test, y_test_pred, heart_df, weight), 0)
rmae_test = round(rmae(y_test, y_test_pred, heart_df, weight), 0)
r2_test = round(r2_score(y_test, y_test_pred, X_test_weights), 4)
print('> evaluation metrics \n')
print('%-10s %20s %10s' % ('metric','training','testing'))
print('%-10s %20s %10s' % ('rmse', rmse_train, rmse_test))
print('%-10s %20s %10s' % ('rmae', rmae_train, rmae_test))
print('%-10s %20s %10s' % ('r2', r2_train, r2_test))
print('\n')
#---- getting feature importance
optimized_rfr_importance = optimized_rfr.best_estimator_.named_steps['rfr'].feature_importances_
indices = np.argsort(-1*optimized_rfr_importance)
rfr_feature_imp_df = pd.DataFrame(optimized_rfr_importance, index=X_train.columns, columns=['importance'])
rfr_feature_imp_df.sort_values(by='importance', ascending=False, inplace=True)
# summarize feature importance
print('> feature importance')
for i in indices:
print('%-8s %-20s' % (round(optimized_rfr_importance[i], 4), f'({features[i]})'))
#---- obtaining results of the grid run
cv_results = optimized_rfr.cv_results_
cv_results_df = pd.DataFrame(cv_results)
print('> hyperparameter tuning results')
print(cv_results_df)
#---- getting best parameters
best_params = optimized_rfr.best_params_
best_score = optimized_rfr.best_score_
print(f'> best hyperparameters = {best_params}')
print(f'> best cv score = {best_score} \n')
#---- saving model results
cv_results_df.to_csv('output\\cv_results.csv', index=False)
best_params_str = ', '.join('{}={}'.format(key, val) for key, val in best_params.items())
with open('output//rfr_results.txt', 'w') as file:
file.write('best parameters = '+best_params_str+'\n')
file.write('rmse: '+'(train='+str(rmse_train)+') (test='+str(rmse_test)+')'+'\n')
file.write('rmae: '+'(train='+str(rmae_train)+') (test='+str(rmae_test)+')'+'\n')
# feature importance plot
plt.style.use('seaborn')
fig, ax = plt.subplots()
ax.barh(range(len(indices)), optimized_rfr_importance[indices], align='center')
ax.set_yticks(range(len(indices)))
ax.set_yticklabels([features[i] for i in indices], fontsize=10)
ax.invert_yaxis()
ax.set_title('Feature Importances', fontsize=22, fontweight='bold')
ax.set_xlabel('Relative Importance', fontsize=12, fontweight='bold')
ax.set_ylabel('Features', fontsize=10, fontweight='bold')
ax.spines['left'].set_color('black')
ax.spines['left'].set_linewidth(2)
ax.spines['bottom'].set_color('black')
ax.spines['bottom'].set_linewidth(2)
ax.grid(True)
fig.savefig('output/feature_importance_plot.png')