-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrf_reg_hp_search.py
163 lines (121 loc) · 6.03 KB
/
rf_reg_hp_search.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
print('importing libs...', end = '\r')
import os
import numpy as np
import pandas as pd
import pickle
import time
import matplotlib.pyplot as plt
import matplotlib.lines as mlines
from matplotlib import cm
import seaborn as sns
from utils_ens import get_Xy_tt
from utils_ens import my_sigmoid
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import RandomForestRegressor
from sklearn import metrics
print('Getting data......', end = '\r')
#pkl_file = open('/home/simon/Documents/Articles/conflict_prediction/data/computerome/currents/rf_selected_features.pkl', 'rb')
pkl_file = open('/home/projects/ku_00017/data/generated/currents/rf_selected_features.pkl', 'rb')
selected_features = pickle.load(pkl_file)
pkl_file.close()
X_train, y_train, X_test, y_test = get_Xy_tt(local = False, binary_y= False)
n_rounds = 50
best_features = selected_features['features'][:9].values # should have specific feature selection for reg... also try with 4..
# hp lists:
max_depth_list = []
n_estimators_list = []
min_samples_split_list = []
criterion_list = []
class_weight_list = []
max_features_list = []
min_samples_leaf_list = []
# metric lists:
# train
train_preds_con = []
test_preds_con = []
train_preds_prob = []
test_preds_prob = []
MSE_train_list = []
MAE_train_list = []
AUC_train_list = []
AP_train_list = []
BS_train_list = []
pr_train_list = []
roc_train_list = []
# test
MSE_test_list = []
MAE_test_list = []
AUC_test_list = []
AP_test_list = []
BS_test_list = []
pr_test_list = []
roc_test_list = []
print('Beginning loop')
for i in range(n_rounds):
# Variable hyper parameters
n_estimators = np.random.randint(100,150) # just to check if it is running at all...
min_samples_split = np.random.randint(2,15) # could go higher..
max_depth = np.random.randint(5,11)
min_samples_leaf = np.random.randint(1,200)
#criterion = ['mse', 'mae'][np.random.randint(0,2)] # mae still not run now with log(y)
criterion ='mse'
max_features = ['auto', 'sqrt', 'log2'][np.random.randint(0,3)]
print(f'defining model {i+1}/{n_rounds}... \nn_estimators: {n_estimators}, \nmin_samples_split: {min_samples_split}, \nmax_depth : {max_depth}, \nmin_sample_leaf : {min_samples_leaf}, \ncriterion : {criterion}, \nmax_features : {max_features} ')
model = RandomForestRegressor( n_estimators=n_estimators, criterion = criterion, max_depth = max_depth,
min_samples_split= min_samples_split, min_samples_leaf= min_samples_leaf,
random_state=i, n_jobs= 18)
print(f'fitting model {i+1}/{n_rounds}...', end = '\r')
model.fit(X_train[best_features], y_train)
# metrics
print(f'getting metrics from model {i+1}/{n_rounds}...', end = '\r')
y_train_pred_con = model.predict(X_train[best_features])
y_test_pred_con = model.predict(X_test[best_features])
y_train_pred_prob = my_sigmoid(y_train_pred_con) # maybe do a logistic regression instead...
y_test_pred_prob = my_sigmoid(y_test_pred_con) # maybe do a logistic regression instead...
y_train_bi = (y_train > 0)*1
y_test_bi = (y_test > 0)*1
# train:
MSE_train_list.append(metrics.mean_squared_error(y_train, y_train_pred_con))
MAE_train_list.append(metrics.mean_absolute_error(y_train, y_train_pred_con))
AUC_train_list.append(metrics.roc_auc_score(y_train_bi, y_train_pred_prob))
AP_train_list.append(metrics.average_precision_score(y_train_bi, y_train_pred_prob))
BS_train_list.append(metrics.brier_score_loss(y_train_bi, y_train_pred_prob))
precision_train, recall_train, _ = metrics.precision_recall_curve(y_train_bi, y_train_pred_prob)
fpr_train, tpr_train, _ = metrics.roc_curve(y_train_bi, y_train_pred_prob)
pr_train_list.append((precision_train, recall_train))
roc_train_list.append((fpr_train, tpr_train))
# test:
MSE_test_list.append(metrics.mean_squared_error(y_test, y_test_pred_con))
MAE_test_list.append(metrics.mean_absolute_error(y_test, y_test_pred_con))
AUC_test_list.append(metrics.roc_auc_score(y_test_bi, y_test_pred_prob))
AP_test_list.append(metrics.average_precision_score(y_test_bi, y_test_pred_prob))
BS_test_list.append(metrics.brier_score_loss(y_test_bi, y_test_pred_prob))
precision_test, recall_test, _ = metrics.precision_recall_curve(y_test_bi, y_test_pred_prob)
fpr_test, tpr_test, _ = metrics.roc_curve(y_test_bi, y_test_pred_prob)
pr_test_list.append((precision_test, recall_test))
roc_test_list.append((fpr_test, tpr_test))
train_preds_con.append(y_train_pred_con)
test_preds_con.append(y_test_pred_con)
train_preds_prob.append(y_train_pred_prob)
test_preds_prob.append(y_test_pred_prob)
# hps:
n_estimators_list.append(n_estimators)
max_depth_list.append(max_depth)
min_samples_split_list.append(min_samples_split)
criterion_list.append(criterion)
max_features_list.append(max_features)
min_samples_leaf_list.append(min_samples_leaf)
loop_string = f'{i+1}/{n_rounds} \n done. MSE test: {MSE_test_list[i]} MSE train: {MSE_train_list[i]} \n MAE test: {MAE_test_list[i]} MAE train: {MAE_train_list[i]} \n AP test: {AP_test_list[i]}, AP train: {AP_train_list[i]} \n'
print(loop_string , end='\r')
print('Making datafame', end = '\r')
hp_df = pd.DataFrame({'n_estimators' : n_estimators_list, 'max_depth' : max_depth_list, 'min_samples_split' : min_samples_split_list,
'min_samples_leaf' : min_samples_leaf_list ,'criterion' : criterion_list, 'max_features' : max_features_list,
'test_preds_con' : test_preds_con, 'test_preds_prob' : test_preds_prob,
'MSE' : MSE_test_list, 'MAE' : MAE_test_list,
'AP' : AP_test_list, 'PR' : pr_test_list, 'ROC' : roc_test_list})
print('Pickling..')
new_file_name = '/home/projects/ku_00017/data/generated/currents/rf_reg_hp_df.pkl'
#new_file_name = '/home/simon/Documents/Articles/conflict_prediction/data/computerome/currents/rf_reg_hp_df.pkl'
output = open(new_file_name, 'wb')
pickle.dump(hp_df, output)
output.close()