-
Notifications
You must be signed in to change notification settings - Fork 1
/
swin.yaml
263 lines (242 loc) · 6.53 KB
/
swin.yaml
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
### base config ###
swin_73var: &73var
#swin hyperparams
nettype: 'swin'
embed_dim: 768
depth: 24
patch_size: 4
window_ratio: 80
num_heads: 8
drop_path_rate: 0.1
full_pos_embed: true
rel_pos: false
mlp_ratio: 4
activation_ckpt: false
checkpointing: false
residual: False
# general
log_to_screen: true
log_to_wandb: true
save_checkpoint: true
project: 'your_wandb_project'
entity: 'your_wandb_entity'
exp_dir: '/pscratch/sd/j/jwillard/FCN_exp/era5_wind/'
# optimizer and other params
optimizer_type: 'adam'
scheduler: 'CosineAnnealingLR'
max_epochs: 70
lr: 1E-3
batch_size: 64
num_data_workers: 8
n_future: 0 # for many future time steps
finetune: false
dt: 1 # time step
add_zenith: false
loss: 'l2'
channel_weights: 'none'
# data
img_size: [720, 1440]
in_channels: [0, 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]
out_channels: [0, 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]
train_data_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/train'
valid_data_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/test'
inf_data_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/out_of_sample'
time_means_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/stats/time_means.npy'
global_means_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/stats/global_means.npy'
global_stds_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/stats/global_stds.npy'
time_diff_means_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/stats/time_diff_means.npy'
time_diff_stds_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/stats/time_diff_stds.npy'
add_orography: False
orography_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/invariants/orog.nc'
add_landmask: False
landmask_path: '/pscratch/sd/p/pharring/73var-6hourly/staging/invariants/lsm.nc'
model_grid_type: "equiangular"
data_loader_config: "dali"
track_channels: ['u10m','v10m','t2m','tcwv']
channel_names:
- u10m
- v10m
- u100m
- v100m
- t2m
- sp
- msl
- tcwv
- u50
- u100
- u150
- u200
- u250
- u300
- u400
- u500
- u600
- u700
- u850
- u925
- u1000
- v50
- v100
- v150
- v200
- v250
- v300
- v400
- v500
- v600
- v700
- v850
- v925
- v1000
- z50
- z100
- z150
- z200
- z250
- z300
- z400
- z500
- z600
- z700
- z850
- z925
- z1000
- t50
- t100
- t150
- t200
- t250
- t300
- t400
- t500
- t600
- t700
- t850
- t925
- t1000
- q50
- q100
- q150
- q200
- q250
- q300
- q400
- q500
- q600
- q700
- q850
- q925
- q1000
swin_73var_short:
<<: *73var
batch_size: 4
max_epochs: 8
log_to_wandb: true
swin_73var_geo: &swin
<<: *73var
loss: 'squared geometric l2'
swin_73var_geo_depth12:
<<: *swin
loss: 'squared geometric l2'
depth: 12
drop_path_rate: 0.1
data_loader_config: "dali"
swin_73var_geo_depth12_chweight:
<<: *swin
loss: 'weighted absolute temp-std squared geometric l2'
channel_weights: 'auto'
depth: 12
drop_path_rate: 0.1
data_loader_config: "dali"
residual: True
add_zenith: True
swin_73var_geo_depth12_chweight_invar: &chwt
<<: *swin
loss: 'weighted absolute temp-std squared geometric l2'
channel_weights: 'auto'
depth: 12
drop_path_rate: 0.1
data_loader_config: "dali"
add_zenith: True
add_orography: True
add_landmask: True
residual: True
swin_73var_geo_depth12_chweight_invar_rel:
<<: *swin
loss: 'weighted relative temp-std squared geometric l2'
channel_weights: 'auto'
depth: 12
drop_path_rate: 0.1
data_loader_config: "dali"
add_zenith: True
add_orography: True
add_landmask: True
residual: True
swin_73var_geo_depth12_chweight_invar_direct:
<<: *swin
loss: 'weighted absolute temp-std squared geometric l2'
channel_weights: 'auto'
depth: 12
drop_path_rate: 0.1
data_loader_config: "dali"
add_zenith: True
add_orography: True
add_landmask: True
swin_73var_geo_depth12_chweight_invar_2step:
<<: *chwt
n_future: 1
max_epochs: 15 # est. 30 min per epoch
lr: 1e-4
finetune: True
pretrained_checkpoint_path: '/pscratch/sd/s/shas1693/results/swin/swin_73var_geo_depth12_chweight_invar/00/training_checkpoints/best_ckpt.tar'
data_loader_config: "pytorch"
swin_73var_geo_depth12_chweight_invar_2step_ep30:
<<: *chwt
n_future: 1
max_epochs: 30 # est. 30 min per epoch
lr: 1e-4
finetune: True
pretrained_checkpoint_path: '/pscratch/sd/s/shas1693/results/swin/swin_73var_geo_depth12_chweight_invar/00/training_checkpoints/best_ckpt.tar'
data_loader_config: "pytorch"
swin_73var_geo_depth24_chweight_invar:
<<: *chwt
depth: 24
swin_73var_geo_depth24_e2048_mlp2_chweight_invar:
<<: *chwt
depth: 24
embed_dim: 2048
mlp_ratio: 2
activation_ckpt: True
data_loader_config: "pytorch"
swin_73var_geo_depth12_chweight_invar_4step:
<<: *chwt
n_future: 3
max_epochs: 15 # est. 11.4hrs total: 2 jobs, 6h30m each
lr: 1e-4
finetune: True
pretrained_checkpoint_path: '/pscratch/sd/s/shas1693/results/swin/swin_73var_geo_depth12_chweight_invar/00/training_checkpoints/best_ckpt.tar'
data_loader_config: "pytorch"
activation_ckpt: True
num_data_workers: 2
swin_73var_geo_depth12_chweight_invar_8step:
<<: *chwt
n_future: 7
max_epochs: 15 # est. 22.6hrs total: 4 jobs, 6h30m each
lr: 1e-4
finetune: True
pretrained_checkpoint_path: '/pscratch/sd/s/shas1693/results/swin/swin_73var_geo_depth12_chweight_invar/00/training_checkpoints/best_ckpt.tar'
data_loader_config: "pytorch"
activation_ckpt: True
num_data_workers: 2
swin_73var_geo_depth12_chweight_invar_8step_int:
<<: *chwt
n_future: 7
batch_size: 4
log_to_wandb: False
max_epochs: 15 # est. 22.6hrs total: 4 jobs, 6h30m each
lr: 1e-4
finetune: True
pretrained_checkpoint_path: '/pscratch/sd/s/shas1693/results/swin/swin_73var_geo_depth12_chweight_invar/00/training_checkpoints/best_ckpt.tar'
data_loader_config: "pytorch"
activation_ckpt: True
num_data_workers: 2