/
Training_candidates.py
225 lines (174 loc) · 9.33 KB
/
Training_candidates.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
221
222
223
224
#!/usr/bin/env python
# coding: utf-8
# In[1]:
#
# Library loading
#
import pandas as pd # manipulate dataframes
import matplotlib
import matplotlib.pyplot as plt # plotting
import numpy as np
np.random.seed = 167 # fix random seed for reproducibility
import time, h5py, imelt, torch
from sklearn.metrics import mean_squared_error
from tqdm import tqdm
# importing shutil module
import shutil
# First we check if CUDA is available
print("CUDA AVAILABLE? ",torch.cuda.is_available())
def get_default_device():
"""Pick GPU if available, else CPU"""
if torch.cuda.is_available():
return torch.device('cuda')
else:
return torch.device('cpu')
device = get_default_device()
print(device)
#
# Loading Data
#
# custom data loader, automatically sent to device
ds = imelt.data_loader("./data/NKAS_viscosity_reference.hdf5",
"./data/NKAS_Raman.hdf5",
"./data/NKAS_density.hdf5",
"./data/NKAS_optical.hdf5",
device)
#
# Training 50 models
#
# Reference architecture
nb_layers = 4
nb_neurons = 300
p_drop = 0.01
nb_exp = 100
for i in range(nb_exp):
print("Training model {}".format(i))
print("...\n")
name = "./model/candidates/l"+str(nb_layers)+"_n"+str(nb_neurons)+"_p"+str(p_drop)+"_m"+str(i)+".pth"
# declaring model
neuralmodel = imelt.model(4,nb_neurons,nb_layers,ds.nb_channels_raman,p_drop=p_drop)
# criterion for match
criterion = torch.nn.MSELoss(reduction='mean')
criterion.to(device) # sending criterion on device
# we initialize the output bias and send the neural net on device
neuralmodel.output_bias_init()
neuralmodel = neuralmodel.float()
neuralmodel.to(device)
optimizer = torch.optim.Adam(neuralmodel.parameters(), lr = 0.0006, weight_decay=0.00) # optimizer
neuralmodel, record_train_loss, record_valid_loss = imelt.training(neuralmodel,ds,
criterion,optimizer,save_switch=True,save_name=name,
train_patience=400,min_delta=0.05,
verbose=True)
print("")
#
# Detect and save the best models
#
#For that we use the global "loss_valid" = loss_viscosity + loss_raman + loss_density + loss_refractiveindex
# scaling coefficients for loss function
# viscosity is always one
# scaling coefficients for loss function
# viscosity is always one
ls = imelt.loss_scales()
entro_scale = ls.entro
raman_scale = ls.raman
density_scale = ls.density
ri_scale = ls.ri
tg_scale = ls.tg
record_loss = pd.DataFrame()
record_loss["name"] = np.zeros(nb_exp)
record_loss["nb_layers"] = np.zeros(nb_exp)
record_loss["nb_neurons"] = np.zeros(nb_exp)
record_loss["loss_ag_train"] = np.zeros(nb_exp)
record_loss["loss_ag_valid"] = np.zeros(nb_exp)
record_loss["loss_am_train"] = np.zeros(nb_exp)
record_loss["loss_am_valid"] = np.zeros(nb_exp)
record_loss["loss_myega_train"] = np.zeros(nb_exp)
record_loss["loss_myega_valid"] = np.zeros(nb_exp)
record_loss["loss_cg_train"] = np.zeros(nb_exp)
record_loss["loss_cg_valid"] = np.zeros(nb_exp)
record_loss["loss_tvf_train"] = np.zeros(nb_exp)
record_loss["loss_tvf_valid"] = np.zeros(nb_exp)
record_loss["loss_Sconf_train"] = np.zeros(nb_exp)
record_loss["loss_Sconf_valid"] = np.zeros(nb_exp)
record_loss["loss_d_train"] = np.zeros(nb_exp)
record_loss["loss_d_valid"] = np.zeros(nb_exp)
record_loss["loss_raman_train"] = np.zeros(nb_exp)
record_loss["loss_raman_valid"] = np.zeros(nb_exp)
record_loss["loss_train"] = np.zeros(nb_exp)
record_loss["loss_valid"] = np.zeros(nb_exp)
for i in range(nb_exp):
name = "./model/candidates/l"+str(nb_layers)+"_n"+str(nb_neurons)+"_p"+str(p_drop)+"_m"+str(i)+".pth"
record_loss.loc[i,"name"] = "l"+str(nb_layers)+"_n"+str(nb_neurons)+"_p"+str(p_drop)+"_m"+str(i)+".pth"
# declaring model
neuralmodel = imelt.model(4,nb_neurons,nb_layers,ds.nb_channels_raman,p_drop=p_drop)
neuralmodel.load_state_dict(torch.load(name, map_location='cpu'))
neuralmodel.to(device)
neuralmodel.eval()
# PREDICTIONS
with torch.set_grad_enabled(False):
# train
y_ag_pred_train = neuralmodel.ag(ds.x_visco_train,ds.T_visco_train)
y_myega_pred_train = neuralmodel.myega(ds.x_visco_train,ds.T_visco_train)
y_am_pred_train = neuralmodel.am(ds.x_visco_train,ds.T_visco_train)
y_cg_pred_train = neuralmodel.cg(ds.x_visco_train,ds.T_visco_train)
y_tvf_pred_train = neuralmodel.tvf(ds.x_visco_train,ds.T_visco_train)
y_raman_pred_train = neuralmodel.raman_pred(ds.x_raman_train)
y_density_pred_train = neuralmodel.density(ds.x_density_train)
y_entro_pred_train = neuralmodel.sctg(ds.x_entro_train)
y_ri_pred_train = neuralmodel.sellmeier(ds.x_ri_train, ds.lbd_ri_train)
# valid
y_ag_pred_valid = neuralmodel.ag(ds.x_visco_valid,ds.T_visco_valid)
y_myega_pred_valid = neuralmodel.myega(ds.x_visco_valid,ds.T_visco_valid)
y_am_pred_valid = neuralmodel.am(ds.x_visco_valid,ds.T_visco_valid)
y_cg_pred_valid = neuralmodel.cg(ds.x_visco_valid,ds.T_visco_valid)
y_tvf_pred_valid = neuralmodel.tvf(ds.x_visco_valid,ds.T_visco_valid)
y_raman_pred_valid = neuralmodel.raman_pred(ds.x_raman_valid)
y_density_pred_valid = neuralmodel.density(ds.x_density_valid)
y_entro_pred_valid = neuralmodel.sctg(ds.x_entro_valid)
y_ri_pred_valid = neuralmodel.sellmeier(ds.x_ri_valid, ds.lbd_ri_valid)
# Compute Loss
# train
record_loss.loc[i,"loss_ag_train"] = np.sqrt(criterion(y_ag_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_myega_train"] = np.sqrt(criterion(y_myega_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_am_train"] = np.sqrt(criterion(y_am_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_cg_train"] = np.sqrt(criterion(y_cg_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_tvf_train"] = np.sqrt(criterion(y_tvf_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_raman_train"] = np.sqrt(criterion(y_raman_pred_train,ds.y_raman_train).item())
record_loss.loc[i,"loss_d_train"] = np.sqrt(criterion(y_density_pred_train,ds.y_density_train).item())
record_loss.loc[i,"loss_Sconf_train"] = np.sqrt(criterion(y_entro_pred_train,ds.y_entro_train).item())
record_loss.loc[i,"loss_ri_train"] = np.sqrt(criterion(y_ri_pred_train,ds.y_ri_train).item())
# validation
record_loss.loc[i,"loss_ag_valid"] = np.sqrt(criterion(y_ag_pred_valid, ds.y_visco_valid).item())
record_loss.loc[i,"loss_myega_valid"] = np.sqrt(criterion(y_myega_pred_valid, ds.y_visco_valid).item())
record_loss.loc[i,"loss_am_valid"] = np.sqrt(criterion(y_am_pred_valid, ds.y_visco_valid).item())
record_loss.loc[i,"loss_cg_valid"] = np.sqrt(criterion(y_cg_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_tvf_valid"] = np.sqrt(criterion(y_tvf_pred_train, ds.y_visco_train).item())
record_loss.loc[i,"loss_raman_valid"] = np.sqrt(criterion(y_raman_pred_valid,ds.y_raman_valid).item())
record_loss.loc[i,"loss_d_valid"] = np.sqrt(criterion(y_density_pred_valid,ds.y_density_valid).item())
record_loss.loc[i,"loss_Sconf_valid"] = np.sqrt(criterion(y_entro_pred_valid,ds.y_entro_valid).item())
record_loss.loc[i,"loss_ri_valid"] = np.sqrt(criterion(y_ri_pred_valid,ds.y_ri_valid).item())
record_loss.loc[i,"loss_train"] = (record_loss.loc[i,"loss_ag_train"] +
record_loss.loc[i,"loss_myega_train"] +
record_loss.loc[i,"loss_am_train"] +
record_loss.loc[i,"loss_cg_train"] +
record_loss.loc[i,"loss_tvf_train"] +
raman_scale*record_loss.loc[i,"loss_raman_train"] +
density_scale*record_loss.loc[i,"loss_d_train"] +
entro_scale*record_loss.loc[i,"loss_Sconf_train"] +
ri_scale*record_loss.loc[i,"loss_ri_train"])
record_loss.loc[i,"loss_valid"] = (record_loss.loc[i,"loss_ag_valid"] +
record_loss.loc[i,"loss_myega_valid"] +
record_loss.loc[i,"loss_am_valid"] +
record_loss.loc[i,"loss_cg_valid"] +
record_loss.loc[i,"loss_tvf_valid"] +
raman_scale*record_loss.loc[i,"loss_raman_valid"] +
density_scale*record_loss.loc[i,"loss_d_valid"] +
entro_scale*record_loss.loc[i,"loss_Sconf_valid"] +
ri_scale*record_loss.loc[i,"loss_ri_valid"])
# Get the 10 best recorded
best_recorded = record_loss.nsmallest(10,"loss_valid")
# Copy the content of
# source to destination
for i in best_recorded.loc[:,"name"]:
shutil.copyfile("./model/candidates/"+i, "./model/best/"+i)
best_recorded.to_csv("./model/best/best_list.csv")