-
Notifications
You must be signed in to change notification settings - Fork 10
/
run_eval.py
210 lines (172 loc) · 7.91 KB
/
run_eval.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
from dl_util import *
from ml_util import *
from chemixnet_util import *
def split_fit_plot_predict(model_arch, X1, X2, Y, vocab, max_len, prefix, dropout=0,\
gate=None, optimizer="adam", lr=0.001, epochs=20,batch_size=32):
X1 = np.array(X1)
X2 = np.array(X2)
Y = np.array(Y, dtype=np.float32)
X1_train, X1_test, y_train, y_test = train_test_split(X1, Y, random_state=1024)
X2_train, X2_test, y_train, y_test = train_test_split(X2, Y, random_state=1024)
model_name = model_arch.__name__
print(model_name)
if "rnn" in model_name:
early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=2)
# model_name
else:
early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=5)
time_start = time.time()
if "tox" in prefix or "hiv" in prefix:
metric = "auc"
model_category = "classification"
else:
metric = "mean_absolute_percentage_error"
model_category = "regression"
if "mlp" in model_name:
if "cnn" in model_name or "rnn" in model_name:#merged architecture
print(gate,model_name)
if gate:#either lstm or gru
model = model_arch(dropout=dropout,optimizer=optimizer,lr=lr, vocab=vocab,\
max_len=max_len, gate=gate )
model_name = model_name.replace("rnn",gate)
else:
print("here")
model = model_arch(dropout=dropout,optimizer=optimizer,lr=lr, vocab=vocab,\
max_len=max_len)
history = model.fit([X1_train,X2_train], y_train, shuffle=True, validation_split=0.1, \
epochs=epochs, batch_size=batch_size, verbose=1, callbacks=[early_stop])
X_test = [X1_test, X2_test]
else:
model = mlp_model(dropout=dropout,optimizer=optimizer,lr=lr)
history = model.fit(X2_train, y_train, shuffle=True, validation_split=0.1, \
epochs=epochs, batch_size=batch_size, verbose=1, callbacks=[early_stop])
X_test = X2_test
else:#just uses SMILES - cnn, rnn, cnn-rnn or bidirectional
if gate:#rnn either gru or lstm
model = model_arch(dropout=dropout,optimizer=optimizer,lr=lr,vocab=vocab,\
max_len=max_len, gate=gate )
model_name = model_name.replace("rnn",gate)
else:
model = model_arch(dropout=dropout,optimizer=optimizer,lr=lr,vocab=vocab,max_len=max_len)
history = model.fit(X1_train, y_train, shuffle=True, validation_split=0.1,\
epochs=epochs, batch_size=batch_size, verbose=1, callbacks=[early_stop])
X_test = X1_test
metrics = model.evaluate(X_test, y_test)
time_end = time.time()
time_elapsed = time_end - time_start
if model_category == "regression":
y_predict = model.predict(X_test).reshape(1,-1)[0]
r2 = r2_score(y_test,y_predict)
mean_squared_err = mse(y_test,y_predict)
mean_absolute_err = mae(y_test, y_predict)
percent_mean_absolute_err = Mape(y_test, y_predict)
mean_absolute_percent_err = metrics[1]
stats = {"mape":percent_mean_absolute_err, "mean_absolute_percent_error": mean_absolute_percent_err, "mae":mean_absolute_err, "mse":mean_squared_err, "r2":r2, "time":time_elapsed}
print("Test mape%:", percent_mean_absolute_err)
else:
loss, accuracy, precision, recall, auc = metrics
stats = { "accuracy":accuracy, "precision":precision, "recall":recall, "auc":auc, "time":time_elapsed}
print("Test AUC:", auc)
if in_jupyter():
plot_history(history, metric)
else:
if time_elapsed >3600:
message = prepare_message(model_name, stats, dropout, epochs, batch_size, lr, \
time_elapsed, ml_task=model_category, prefix=prefix)
subject = prefix+"_"+model_name+"_dropout_"+str(dropout)+"_epochs_"\
+str(epochs)+"_"+str(batch_size)+"_"+str(lr)
try:
send_email(subject, message)
except:
print("Unable to send email")
file_suffix = prefix+"_"+model_name+"_dropout_"+str(dropout)\
+"_epochs_"+str(epochs)+"_batch_"+str(batch_size)+"_lr_"+str(lr)
save_history(history, "history_"+file_suffix, "model")
saveData(stats,"stats_"+ file_suffix, "model")
print("Stats saved in model/stats_"+ file_suffix)
return y_test,y_predict
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--model", help= "which model to train", required=True)
parser.add_argument("-s", "--dataset", help= "dataset to use", required=True)
parser.add_argument("-f", "--fingerprint", help= "which fingerprint - if no fingerprint, it's maccs", required=False)
parser.add_argument("-o", "--optimizer", help= "which optimizer(default is adam)", required=False)
parser.add_argument("-d", "--dropout", help= "amount of dropout", required=False)
parser.add_argument("-l", "--layers", help= "number of layers", required=False)
parser.add_argument("-e", "--epochs", help= "epochs", required=False)
parser.add_argument("-b", "--batch_size", help= "size of batch", required=False)
parser.add_argument("-r", "--learning_rate", help= "learning rate", required=False)
parser.add_argument("-c", "--recurrent_connections", help= "default is 100", required=False)
args = parser.parse_args()
model_type = args.model
dataset = args.dataset
if "gru" in model_type:
gate = "gru"
model_type = model_type.replace("gru","rnn")
elif "lstm" in model_type:
gate = "lstm"
model_type = model_type.replace("lstm","rnn")
else:
gate = None
if dataset == "tox" or dataset == "hiv":
X1 = loadNumpy(dataset+'_sequences')
X2 = loadNumpy(dataset+'_maccs' )
if "tox" in dataset:
Y = loadNumpy('tox_nontoxic')
vocab_size, max_len = 42, 940
else:#hiv
Y = loadNumpy('hiv_active')
vocab_size, max_len = 54, 400
elif "esol" in dataset:
X1 = loadNumpy('esol_sequences')
X2 = loadNumpy('esol_maccs' )
if "standardized" in dataset:
Y = loadNumpy('esol_standardized_solubility')#standardized data
else:
Y = loadNumpy('esol_solubility')
vocab_size, max_len = 33, 98
elif "opv" in dataset:
if "exp" in dataset:
X1 = loadNumpy('opv_exp_sequences')
X2 = loadNumpy('opv_exp_maccs')
Y = loadNumpy('opv_exp_homo')
vocab_size, max_len = 32,176
else:
X1 = loadNumpy('opv_dft_sequences')
X2 = loadNumpy('opv_dft_maccs')
vocab_size, max_len = 31, 186
dft_type = dataset.split("opv_")[1]
Y = loadNumpy('opv_'+dft_type+'_homo')
else:
print("Dataset not defined")
exit(4)
if args.fingerprint:
fp_type = args.fingerprint
else:
fp_type = "maccs"
if args.dropout:
dropout = float(args.dropout)
else:
dropout =0
if args.epochs:
epochs = int(args.epochs)
else:
epochs = 20
if args.optimizer:
optimizer = args.optimizer
else:
optimizer = "adam"
if args.batch_size:
batch_size = int(args.batch_size)
else:
batch_size = 32
if args.learning_rate:
lr = float(args.learning_rate)
else:
lr = 0.001
if args.recurrent_connections:
recur_conn = int(args.recurrent_connections)
else:
recur_conn = 100
split_fit_plot_predict(eval(model_type+"_model"), X1, X2, Y, vocab_size, max_len, args.dataset,dropout=dropout, optimizer=optimizer, lr=lr, epochs=epochs,batch_size=batch_size, gate=gate)