-
Notifications
You must be signed in to change notification settings - Fork 28
/
ex3_darcy_inv.py
163 lines (139 loc) · 6.64 KB
/
ex3_darcy_inv.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
from libs_path import *
from libs import *
def main():
with open(os.path.join(SRC_ROOT, 'config.yml')) as f:
config = yaml.full_load(f)
test_name = os.path.basename(__file__).split('.')[0]
config = config[test_name]
args = get_args_2d(**config)
cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device('cuda' if cuda else 'cpu')
kwargs = {'pin_memory': True} if cuda else {}
get_seed(args.seed)
train_path = os.path.join(DATA_PATH, 'piececonst_r421_N1024_smooth1.mat')
test_path = os.path.join(DATA_PATH, 'piececonst_r421_N1024_smooth2.mat')
train_dataset = DarcyDataset(data_path=train_path,
subsample_attn=args.subsample_attn,
subsample_nodes=args.subsample_nodes,
subsample_inverse=args.subsample_attn,
subsample_method='average',
inverse_problem=True,
train_data=True,
noise=args.noise,
train_len=1024,)
bsz = 2 if args.subsample_attn <=7 else args.batch_size
train_loader = DataLoader(train_dataset, batch_size=bsz, shuffle=True,
drop_last=True, **kwargs)
valid_dataset = DarcyDataset(data_path=test_path,
normalizer_x=train_dataset.normalizer_x,
subsample_attn=args.subsample_attn,
subsample_nodes=args.subsample_nodes,
subsample_inverse=args.subsample_attn,
subsample_method='average',
inverse_problem=True,
train_data=False,
noise=args.noise,
valid_len=100,)
valid_loader = DataLoader(valid_dataset, batch_size=args.val_batch_size, shuffle=False,
drop_last=False, **kwargs)
n_grid = int(((421 - 1)/args.subsample_nodes) + 1)
n_grid_c = int(((421 - 1)/args.subsample_attn) + 1)
downsample, _ = DarcyDataset.get_scaler_sizes(n_grid, n_grid_c)
sample = next(iter(train_loader))
print('='*20, 'Data loader batch', '='*20)
for key in sample.keys():
print(key, "\t", sample[key].shape)
print('='*(40 + len('Data loader batch')+2))
if is_interactive():
idx = 3
u = sample['node']
a = sample['target']
elem = train_dataset.elem
node = train_dataset.pos
ah = a[..., 0]
uh = train_dataset.normalizer_x.inverse_transform(u)
uh = F.interpolate(uh[..., 0].unsqueeze(1), size=(
n_grid_c, n_grid_c), mode='bilinear', align_corners=True)
ah = ah[idx].numpy().reshape(-1)
uh = uh[idx].numpy().reshape(-1)
showsolution(node, elem, uh, width=600, height=500)
fig, ax = plt.subplots(figsize=(10, 10))
ah_plot = ax.imshow(ah.reshape(n_grid_c, n_grid_c), cmap='RdBu')
fig.colorbar(ah_plot, ax=ax, anchor=(0, 0.3), shrink=0.8)
config['upscaler_size'] = (n_grid_c, n_grid_c), (n_grid_c, n_grid_c)
config['normalizer'] = train_dataset.normalizer_y.to(device)
config['downscaler_size'] = downsample
config['attn_norm'] = not args.layer_norm
for arg in vars(args):
if arg in config.keys():
config[arg] = getattr(args, arg)
torch.manual_seed(seed=args.seed)
torch.cuda.manual_seed(seed=args.seed)
torch.cuda.empty_cache()
model = FourierTransformer2D(**config)
model = model.to(device)
print(
f"\nModel: {model.__name__}\t Number of params: {get_num_params(model)}")
n_head = config['n_head']
model_name, result_name = get_model_name(model='darcy',
num_encoder_layers=config['num_encoder_layers'],
n_hidden=config['n_hidden'],
attention_type=config['attention_type'],
layer_norm=config['layer_norm'],
grid_size=n_grid,
inverse_problem=True,
additional_str=f'{n_head}h_{args.noise:.1e}'
)
print(f"Saving model and result in {MODEL_PATH}/{model_name}\n")
epochs = args.epochs
tqdm_mode = 'epoch' if not args.show_batch else 'batch'
lr = args.lr
h = 1/n_grid_c
optimizer = torch.optim.Adam(model.parameters(), lr=lr)
scheduler = OneCycleLR(optimizer, max_lr=lr,
div_factor=1e4,
final_div_factor=1e4,
pct_start=0.3,
steps_per_epoch=len(train_loader), epochs=epochs)
loss_func = WeightedL2Loss2d(regularizer=False, h=h)
metric_func = WeightedL2Loss2d(regularizer=False, h=h)
result = run_train(model, loss_func, metric_func,
train_loader, valid_loader,
optimizer, scheduler,
train_batch=train_batch_darcy,
validate_epoch=validate_epoch_darcy,
epochs=epochs,
patience=None,
tqdm_mode=tqdm_mode,
model_name=model_name,
result_name=result_name,
device=device)
model.load_state_dict(torch.load(os.path.join(MODEL_PATH, model_name)))
model.eval()
val_metric = validate_epoch_darcy(model, metric_func, valid_loader, device)
print(f"\nBest model's validation metric in this run: {val_metric}")
plt.figure(1)
loss_train = result['loss_train']
loss_val = result['loss_val']
plt.semilogy(loss_train[:, 0], label='train')
plt.semilogy(loss_val, label='valid')
plt.grid(True, which="both", ls="--")
plt.legend()
plt.show()
sample = next(iter(valid_loader))
u = sample['node']
pos = sample['pos']
a = sample['target']
grid = sample['grid']
with torch.no_grad():
model.eval()
_out = model(u.to(device), None, pos.to(device), grid.to(device))
preds = _out['preds']
for i in range(args.val_batch_size):
z = preds[i, ..., 0].cpu().numpy()
z_true = a[i, ..., 0].cpu().numpy()
_ = showcontour(z, width=500, height=500,)
_ = showcontour(z_true, width=500, height=500,)
print("Relative error: ", np.linalg.norm(z-z_true)/np.linalg.norm(z_true))
if __name__ == '__main__':
main()