AutoML-Kaggle-Competition/version2.py at master · leihchen/AutoML-Kaggle-Competition · GitHub

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
# testing xgboost, submit GDBT w/o grid search
import sklearn
import sklearn.svm as svm
import sklearn.tree as tree
import sklearn.ensemble as ensemble
import sklearn.decomposition as decomposition
import sklearn.preprocessing as preprocessing
import sklearn.model_selection as model_selection
import sklearn as sk
import numpy as np
import pandas as pd
import xgboost as xgb
import matplotlib.pyplot as plt
from xgboost import plot_importance
from tqdm import tqdm
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import GridSearchCV
from parser_1 import apply_flops, apply_ops_hist, apply_init_params, diff_avg
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)


def test_to_csv(Regr_train_err, Regr_validation_err, Data_train_std, Data_validation_std, filename='test_result.csv'):
    pt = Regr_train_err.predict(Data_train_std)
    pv = Regr_validation_err.predict(Data_validation_std)
    df_test = pd.read_csv("data/test.csv")
    df_ex = pd.DataFrame(
        {'id': ['test_' + str(i // 2) + '_val_error' if i % 2 == 0 else 'test_' + str(i // 2) + '_train_error'
                for i in range(df_test.shape[0] * 2)],
         'Predicted': [pv[i // 2] if i % 2 == 0 else pt[i // 2]
                       for i in range(df_test.shape[0] * 2)]})
    df_ex.to_csv(filename, header=True, index=False)


df = pd.read_csv("data/train-1185.csv")
D = df.shape[0]
feature_all = list(df.columns)
# for i in range(len(feature_all)):
#     print(i, feature_all[i])
# flops = apply_flops(df)
# flops.to_csv('flops.csv', index=None)
flops = pd.read_csv('flops.csv')
y_val = df["val_error"]
y_tr = df["train_error"]

X = df[feature_all[17:] + ['number_parameters', 'epochs']].\
    join(flops).\
    join(apply_ops_hist(df)).\
    join(apply_init_params(df))\
    .\
    join(diff_avg(df[feature_all[17:67]], 'val_accs_diff')).\
    join(diff_avg(df[feature_all[67:117]], 'val_losses_diff')).\
    join(diff_avg(df[feature_all[117:167]], 'train_accs_diff')).\
    join(diff_avg(df[feature_all[167:217]], 'train_losses_diff'))  # train acc
pd.DataFrame(X).to_csv('data matrix.csv')
model_features = X.columns
X = preprocessing.scale(X)

X_train_tr, X_test_tr, y_train_tr, y_test_tr = model_selection.train_test_split(X, y_tr)
# regr_tr = sk.linear_model.LinearRegression()
# regr_tr.fit(X_train_tr, y_train_tr)
# y_pred_tr = regr_tr.predict(X_test_tr)
# print('Linear regression training err: R2 metric = ', sk.metrics.r2_score(y_test_tr, y_pred_tr))
parameter_candidates = {
    "subsample":[0.5, 0.55, 0.6, 0.65, 0.7]
}


# regr_tr = GridSearchCV(estimator=xgb.XGBRegressor(objective ='reg:logistic',missing=np.nan, max_depth=4, learning_rate=0.08, n_estimators=190, subsample=0.5),
#                     param_grid=parameter_candidates,
#                     cv=5,
#                     refit=True,
#                     error_score=0,
#                     n_jobs=-1)
# regr_tr = sk.linear_model.LinearRegression()
# regr_tr.fit(X_train_tr, y_train_tr)  # cross validation
# regr_tr = ensemble.GradientBoostingRegressor(loss="ls", max_depth=4, subsample=0.9, n_estimators=140)
regr_tr = xgb.XGBRegressor(objective ='reg:logistic',missing=np.nan, max_depth=4, learning_rate=0.08, n_estimators=190, subsample=0.5)

regr_tr.fit(X, y_tr)  # submission
y_pred_tr = regr_tr.predict(X_test_tr)
# print('the optimal params we found are: ', regr_tr.best_params_)
print('training err: R2 metric = ', sk.metrics.r2_score(y_test_tr, y_pred_tr))

#############
# Use same X: Yes
# X = df[feature_all[57:67] + ['number_parameters', 'epochs']].join(flops) # val acc
# X = preprocessing.scale(X)
# only 1 feature survived pca 95% var, acc decreases?
# print(X.shape)
# pca = decomposition.PCA(.95)
# X = pca.fit_transform(X)
# print(X.shape)

X_train_val, X_test_val, y_train_val, y_test_val = model_selection.train_test_split(X, y_val)
# regr_val = sk.linear_model.LinearRegression()
# regr_val.fit(X_train_val, y_train_val)
# y_pred_val = regr_val.predict(X_test_val)
# print('Linear regression benchmark validation err: R2 metric = ', sk.metrics.r2_score(y_test_val, y_pred_val))

###

# regr_val = GridSearchCV(estimator=xgb.XGBRegressor(objective ='reg:logistic',missing=np.nan, max_depth=4,learning_rate=0.08, n_estimators=200, subsample=0.7),
#                     param_grid=parameter_candidates,
#                     cv=5,
#                     refit=True,
#                     error_score=0,
#                     n_jobs=-1)
# regr_val = ensemble.GradientBoostingRegressor(loss="huber", max_depth=5, subsample=1.0)
regr_val = xgb.XGBRegressor(objective ='reg:logistic',missing=np.nan, max_depth=4,learning_rate=0.08, n_estimators=200, subsample=0.7)

# regr_val = sk.linear_model.LinearRegression()
# regr_val.fit(X_train_val, y_train_val)  # cross validation
regr_val.fit(X, y_val)  # submission
y_pred_val = regr_val.predict(X_test_val)
# print('the optimal params we found are: ', regr_val.best_params_)
print('validation err: R2 metric = ', sk.metrics.r2_score(y_test_val, y_pred_val))

##### CV
# scores = cross_val_score(regr_tr, X, y_tr)
# print("training Accuracy: %0.3f (+/- %0.3f)" % (scores.mean(), scores.std() * 2))
# scores = cross_val_score(regr_val, X, y_val)
# print("validation Accuracy: %0.3f (+/- %0.3f)" % (scores.mean(), scores.std() * 2))


# # # ######
df_t = pd.read_csv("data/test.csv")
# for i in range(len(df_t.columns)):
#     print(i, df_t.columns[i])
flops_t = apply_flops(df_t)
op_hist = apply_ops_hist(df_t)
# X_ex_tr = preprocessing.scale(df_t[list(df_t.columns[153:163]) + ['number_parameters', 'epochs']].join(flops_t))
X_ex = preprocessing.scale(df_t[feature_all[17:] + ['number_parameters', 'epochs']].
                           join(flops_t).
                           join(apply_ops_hist(df_t)).
                           join(apply_init_params(df_t)).
                           join(diff_avg(df_t[feature_all[17:67]], 'val_accs_diff')).\
    join(diff_avg(df_t[feature_all[67:117]], 'val_losses_diff')).\
    join(diff_avg(df_t[feature_all[117:167]], 'train_accs_diff')).\
    join(diff_avg(df_t[feature_all[167:217]], 'train_losses_diff')))
test_to_csv(regr_tr, regr_val, X_ex, X_ex, filename="v19_2.csv")


# indices = np.argsort(regr_tr.feature_importances_)[-15:]
# # plt.figure(figsize=[10, 15], dpi=200)
# plt.title('Feature Importances')
# plt.barh(range(len(indices)), regr_tr.feature_importances_[indices], color='b', align='center')
# plt.yticks(range(len(indices)), [model_features[i] for i in indices], fontsize=5)
# ### explore xgboost
# # xgb_train = xgb.DMatrix(X_train_val, label=y_train_val)
# # xgb_test = xgb.DMatrix(X_test_val, label=y_test_val)
# # param = {'max_depth': 5, 'silent': 1, 'objective': 'reg:squarederror'}
# # evallist = [(xgb_test, 'eval'), (xgb_train, 'train')]
# # num_round = 100
# # bst = xgb.train(param, xgb_train, num_round, evallist)
# # y_pred_val = bst.predict(xgb_test)
# # print('validation err: R2 metric = ', sk.metrics.r2_score(y_test_val, y_pred_val))
# # fig, ax = plt.subplots(figsize=(10, 15))
# # plot_importance(bst, height=0.5, max_num_features=64, ax=ax)
# plt.savefig('tuned_model_tr_imp.png', dpi=300)
# # plt.show()