-
Notifications
You must be signed in to change notification settings - Fork 1
/
pinn_code.py
174 lines (156 loc) · 6.06 KB
/
pinn_code.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
# %%
import numpy as np
import tensorflow as tf
from tensorflow import keras
from time import time
from physics.initial import *
from models.model import *
from utils.plot import *
from physics.equation import *
from config import *
from tqdm import tqdm
# %%
DTYPE = 'float32'
tf.keras.backend.set_floatx(DTYPE)
# Training loop
# Nb de boucle du NN, modele, Valeurs du dataset, CI de X, ..., fonction reelle, N,dim,
# reduit le calcul par petit groupe, barre de progression, ratio de data utilisé
# pour validation/test/entrainement
def train(epochs, pinn, X_r, X_data, u_data, f_real, N, dimension, batch_size, render_bar=True, val_ratio=0.1, begin_from=80000):
losses = []
val_losses = []
t0 = time()
# Setting validation dataset size
total_size = tf.shape(X_r)[0].numpy()
val_size = (val_ratio * total_size).astype(int)
if batch_size > total_size - val_size:
print(total_size - val_size)
raise ValueError("Batch size > train size")
# Setting train and validation data
X_data_test = tf.concat(X_data, axis=0)[-val_size:]
X_test = tf.concat([X_r[-val_size:], X_data_test], axis=0)
X_r, X_data, u_data = X_r[:-val_size], [x[:-val_size]
for x in X_data], [x[:-val_size] for x in u_data]
num_steps = (np.ceil(tf.shape(X_r)[0].numpy())/batch_size).astype(int)
val_loss = pinn.test_step(X_test, f_real)
val_losses.append(val_loss)
if render_bar:
progress_bar = tqdm(range(epochs+1))
else:
progress_bar = range(epochs+1)
# Main training loop over epochs
for i in progress_bar:
loss = 0
# Loop over all batches
for j in range(num_steps):
idx_i = j*batch_size
idx_e = (j+1)*batch_size
X_rj, X_dataj, u_dataj = X_r[idx_i:idx_e], tf.constant(np.array(X_data)[:, idx_i:idx_e]),\
tf.constant(np.array(u_data)[:, idx_i:idx_e])
loss_i, loss_b1, loss_b2, loss_r, lambda_b, lambda_bv, lambda_r = \
pinn.train_step(X_rj, X_dataj, u_dataj,
i) # Calling train step on batch
loss_j = loss_i + loss_b1 + loss_b2 + loss_r
loss += loss_j
loss = loss / num_steps
losses.append(loss)
# Rendering training metrics
if render_bar:
progress_bar.set_description(f"Epoch {i}: Loss= {loss}")
if (i+1) % 50 == 0:
val_loss = pinn.test_step(X_test, f_real)
val_losses.append(val_loss)
#print(f"Epoch {i}: val_loss : {val_loss}")
if (i+1) % 10000 == 0:
print(f'It {i}: residual_loss = {loss_r}\
| initial_loss = {loss_i}\
| boundary_loss_x = {loss_b1}\
| boundary_loss_v = {loss_b2}\
| lambda_b = {lambda_b}\
| lambda_bv = {lambda_bv}\
| lambda_r = {lambda_r}')
if (i+1) % 1000 == 0:
# Change name file for another train
pinn.model.save_weights('results/pinn200.h5')
if (i+1) % 2000 == 0:
if dimension == 1:
plot1dgrid(lb, ub, N, pinn.model, i+begin_from)
if (i+1) % 50 == 0:
plot_curve(i, losses, val_losses, 'plot2.png')
print('\nComputation time: {} seconds'.format(time()-t0))
return losses, val_losses
# Multiple trainings for different number of points
def multi_train():
times = []
points = np.concatenate(
(np.arange(0, 100, 10), np.arange(100, 1050, 50)), axis=0
)
for N_0 in points:
config = define_config()
c, a, dimension, tmin, tmax, xmin, xmax, N_b, N_r, N_0, lr, epochs = \
config['c'],\
config['a'],\
config['dimension'],\
config['tmin'],\
config['tmax'],\
config['xmin'],\
config['xmax'],\
N_0,\
N_0,\
N_0,\
config['learning_rate'],\
config['epochs']
X_data, u_data, time_x, X_r = set_training_data(
tmin, tmax, xmin, xmax, dimension, N_0, N_b, N_r)
bound1 = [tmin] + [xmin for _ in range(dimension)]
bound2 = [tmax] + [xmax for _ in range(dimension)]
lb, ub = tf.constant(bound1, dtype=DTYPE), tf.constant(
bound2, dtype=DTYPE)
opt = keras.optimizers.Adam(learning_rate=lr)
pinn = PINN(dimension+1, 1, dimension, ub, lb, c)
pinn.compile(opt)
t_0 = time()
train(epochs, pinn, X_r, X_data, u_data, true_u,
N=100, dimension=dimension, batch_size=32)
times.append(time()-t_0)
return times
# %%
if __name__ == '__main__':
config = define_config()
c, a, dimension, tmin, tmax, xmin, xmax, N_b, N_r, N_0, lr, epochs = \
config['c'],\
config['a'],\
config['dimension'],\
config['tmin'],\
config['tmax'],\
config['xmin'],\
config['xmax'],\
config['N_b'],\
config['N_r'],\
config['N_0'],\
config['learning_rate'],\
config['epochs']
X_data, u_data, time_x, X_r = set_training_data(
tmin, tmax, xmin, xmax, dimension, N_0, N_b, N_r)
plot_training_points(dimension, time_x)
bound1 = [tmin] + [xmin for _ in range(dimension)]
bound2 = [tmax] + [xmax for _ in range(dimension)]
lb, ub = tf.constant(bound1, dtype=DTYPE), tf.constant(bound2, dtype=DTYPE)
plot1dgrid_real(lb, ub, 200, lambda x: true_u(x, a, c), 999999)
opt = keras.optimizers.Adam(learning_rate=lr)
hist = []
pinn = PINN(dimension+1, 1, dimension, ub, lb, c)
pinn.compile(opt)
pinn.model.load_weights('results/pinn100.h5')
batch_size_max = int(0.9*N_b) # 30% of train dataset
train(epochs, pinn, X_r, X_data, u_data, true_u, N=100,
dimension=dimension, batch_size=batch_size_max)
# Test
model = pinn.model
N = 70
fps = 5
tspace = np.linspace(lb[0], ub[0], N + 1)
plot1d(lb, ub, N, tspace, model, fps)
N = 100
tspace = np.linspace(lb[0], ub[0], N + 1)
plot1dgrid(lb, ub, N, model, 0)