-
Notifications
You must be signed in to change notification settings - Fork 0
/
gravity_wave_model.py
360 lines (303 loc) · 12.5 KB
/
gravity_wave_model.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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
import torch
import torch.nn as nn
import importlib
model = importlib.import_module('Prithvi-WxC.PrithviWxC.model')
from distributed import print0
torch.set_float32_matmul_precision("high")
class Encoder(nn.Module):
"""Encoder, consisting of a series of convolutional layers
with batch normalization and ReLU activation.
Args:
in_channels (int): Number of input channels.
hidden_channels (int): Number of channels in the hidden layers.
"""
def __init__(self, in_channels, hidden_channels):
super(Encoder, self).__init__()
# First encoding block
self.encoder1 = nn.Sequential(
nn.Conv2d(in_channels, hidden_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(hidden_channels),
nn.ReLU(inplace=True),
nn.Conv2d(hidden_channels, hidden_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(hidden_channels),
nn.ReLU(inplace=True),
)
# Second encoding block
self.encoder2 = nn.Sequential(
nn.Conv2d(hidden_channels, hidden_channels * 2, kernel_size=3, padding=1),
nn.BatchNorm2d(hidden_channels * 2),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 2, hidden_channels * 2, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 2),
nn.ReLU(inplace=True),
)
# Third encoding block
self.encoder3 = nn.Sequential(
nn.Conv2d(
hidden_channels * 2, hidden_channels * 4, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 4),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 4, hidden_channels * 4, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 4),
nn.ReLU(inplace=True),
)
# Fourth encoding block
self.encoder4 = nn.Sequential(
nn.Conv2d(
hidden_channels * 4, hidden_channels * 8, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 8),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 8, hidden_channels * 8, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 8),
nn.ReLU(inplace=True),
)
def forward(self, x):
"""Forward pass for the encoder.
Args:
x (torch.Tensor): Input tensor.
Returns:
Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: Encoded tensors from each layer.
"""
enc1 = self.encoder1(x)
enc2 = self.encoder2(enc1)
enc3 = self.encoder3(enc2)
enc4 = self.encoder4(enc3)
return enc1, enc2, enc3, enc4
class Decoder(nn.Module):
"""Decoder for UNet, consisting of convolutional layers to upsample and reconstruct the original input size.
Args:
hidden_channels (int): Number of hidden channels in the decoder layers.
out_channels (int): Number of output channels.
"""
def __init__(self, hidden_channels, out_channels):
super(Decoder, self).__init__()
# Fourth decoding block
self.decoder4 = nn.Sequential(
nn.Conv2d(
hidden_channels * 16, hidden_channels * 8, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 8),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 8, hidden_channels * 8, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 8),
nn.ReLU(inplace=True),
)
# Third decoding block
self.decoder3 = nn.Sequential(
nn.Conv2d(
hidden_channels * 12, hidden_channels * 4, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 4),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 4, hidden_channels * 4, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 4),
nn.ReLU(inplace=True),
)
# Second decoding block
self.decoder2 = nn.Sequential(
nn.Conv2d(
hidden_channels * 6, hidden_channels * 2, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 2),
nn.ReLU(inplace=True),
nn.Conv2d(
hidden_channels * 2, hidden_channels * 2, kernel_size=3, padding=1
),
nn.BatchNorm2d(hidden_channels * 2),
nn.ReLU(inplace=True),
)
# First decoding block
self.decoder1 = nn.Sequential(
nn.Conv2d(hidden_channels * 3, hidden_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(hidden_channels),
nn.ReLU(inplace=True),
nn.Conv2d(hidden_channels, hidden_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(hidden_channels),
nn.ReLU(inplace=True),
)
# Final output layer
self.final_conv = nn.Conv2d(hidden_channels, out_channels, kernel_size=1)
def forward(self, enc1, enc2, enc3, enc4, bottleneck):
"""Forward pass for the decoder, concatenating the encoder outputs with the bottleneck.
Args:
enc1, enc2, enc3, enc4 (torch.Tensor): Encoder outputs.
bottleneck (torch.Tensor): Bottleneck tensor from the transformer.
Returns:
torch.Tensor: Final output tensor.
"""
dec4 = torch.cat((bottleneck, enc4), dim=1)
dec4 = self.decoder4(dec4)
dec3 = torch.cat((dec4, enc3), dim=1)
dec3 = self.decoder3(dec3)
dec2 = torch.cat((dec3, enc2), dim=1)
dec2 = self.decoder2(dec2)
dec1 = torch.cat((dec2, enc1), dim=1)
dec1 = self.decoder1(dec1)
output = self.final_conv(dec1)
return output
class UNetWithTransformer(nn.Module):
"""UNet model with a transformer-based bottleneck for climate data processing.
Args:
lr (float): Learning rate for the optimizer.
in_channels (int): Number of input channels.
hidden_channels (int): Number of hidden channels.
out_channels (int): Number of output channels.
n_lats_px (int): Number of latitude pixels.
n_lons_px (int): Number of longitude pixels.
in_channels_static (int): Number of static input channels.
mask_unit_size_px (list[int]): Size of masking units for the transformer.
patch_size_px (list[int]): Size of patches for the transformer.
device (str): Device to run the model on.
ckpt_singular (str): Path to the checkpoint for pre-trained weights.
"""
def __init__(
self,
lr: float = 1e-3,
in_channels: int = 488,
hidden_channels: int = 160,
out_channels: int = 366,
n_lats_px: int = 64,
n_lons_px: int = 128,
in_channels_static: int = 3,
mask_unit_size_px: list[int] = [8, 16],
patch_size_px: list[int] = [2, 2],
device="cpu",
ckpt_singular=None,
):
super().__init__()
self.lr: float = lr
self.patch_size_px: list[int] = patch_size_px
self.out_channels: int = out_channels
self.encoder = Encoder(in_channels, hidden_channels)
self.decoder = Decoder(hidden_channels, out_channels)
# Transformer model setup using PrithviWxC
kwargs = {
"in_channels": 1280,
"input_size_time": 1,
"n_lats_px": 64,
"n_lons_px": 128,
"patch_size_px": [2, 2],
"mask_unit_size_px": [8, 16],
"mask_ratio_inputs": 0.5,
"embed_dim": 2560,
"n_blocks_encoder": 12,
"n_blocks_decoder": 2,
"mlp_multiplier": 4,
"n_heads": 16,
"dropout": 0.0,
"drop_path": 0.05,
"parameter_dropout": 0.0,
"residual": "none",
"masking_mode": "both",
"decoder_shifting": False,
"positional_encoding": "absolute",
"checkpoint_encoder": [3, 6, 9, 12, 15, 18, 21, 24],
"checkpoint_decoder": [1, 3],
"in_channels_static": 3,
"input_scalers_mu": torch.tensor([0] * 1280),
"input_scalers_sigma": torch.tensor([1] * 1280),
"input_scalers_epsilon": 0,
"static_input_scalers_mu": torch.tensor([0] * 3),
"static_input_scalers_sigma": torch.tensor([1] * 3),
"static_input_scalers_epsilon": 0,
"output_scalers": torch.tensor([0] * 1280),
}
self.model = model.PrithviWxC(**kwargs)
# Freeze transformer model parameters
for param in self.model.parameters():
param.requires_grad = False
# Load pre-trained weights if checkpoint is provided
if ckpt_singular:
print0(f"Starting to load model from {ckpt_singular}")
state_dict = torch.load(
f=ckpt_singular, map_location=device, weights_only=True
)
# Compare the keys in model and saved state_dict
model_keys = set(self.model.state_dict().keys())
saved_state_dict_keys = set(state_dict.keys())
# Find keys that are in the model but not in the saved state_dict
missing_in_saved = model_keys - saved_state_dict_keys
# Find keys that are in the saved state_dict but not in the model
missing_in_model = saved_state_dict_keys - model_keys
# Find keys that are common between the model and the saved state_dict
common_keys = model_keys & saved_state_dict_keys
# Print the common keys
if common_keys:
print0(f"Keys loaded : {common_keys}")
# Print the discrepancies
if missing_in_saved:
print0(f"Keys present in model but missing in saved state_dict: {missing_in_saved}")
if missing_in_model:
print0(f"Keys present in saved state_dict but missing in model: {missing_in_model}")
# Load the state_dict with strict=False to allow partial loading
self.model.load_state_dict(state_dict=state_dict, strict=False)
print0('=>'*10, f"Model loaded from {ckpt_singular}...")
print0("Loaded weights")
def forward(self, batch: dict[str, torch.Tensor]) -> torch.Tensor:
"""Forward pass of the model.
Args:
batch (dict[str, torch.Tensor]): Dictionary containing input data, lead time, and static inputs.
Returns:
torch.Tensor: The final output of the model.
"""
x = batch["x"]
lead_time = batch["lead_time"]
static = batch["static"]
x = x.squeeze(1)
# Encode input
enc1, enc2, enc3, enc4 = self.encoder(x)
# Reshape encoded data for the transformer
batch_size, c, h, w = enc4.size()
enc4_reshaped = enc4.unsqueeze(1)
# Prepare input for transformer model
batch_dict = {
"x": enc4_reshaped,
"y": enc4,
"lead_time": lead_time,
"static": static,
"input_time": torch.zeros_like(lead_time),
}
# Transformer forward pass
transformer_output = self.model(batch_dict)
transformer_output_reshaped = transformer_output.view(batch_size, c, h, w)
# Decode the transformer output
output = self.decoder(enc1, enc2, enc3, enc4, transformer_output_reshaped)
return output
def configure_optimizers(self):
"""Configure the optimizer for training."""
optimizer = torch.optim.Adam(self.parameters(), lr=self.lr)
return optimizer
def validation_step(
self, batch: dict[str, torch.Tensor], batch_idx: int = None
) -> torch.Tensor:
"""Perform a validation step.
Args:
batch (dict[str, torch.Tensor]): Input batch for validation.
batch_idx (int, optional): Batch index.
Returns:
torch.Tensor: Validation loss.
"""
y_hat: torch.Tensor = self(batch)
# Compute loss
loss: torch.Tensor = torch.nn.functional.mse_loss(
input=y_hat, target=batch["target"]
)
return loss
def get_model(self):
"""Return the encoder, decoder, and transformer model.
Returns:
Tuple[nn.Module, nn.Module, nn.Module]: The transformer, decoder, and encoder.
"""
return self.model, self.decoder, self.encoder