forked from USGS-R/drb-inland-salinity-ml
-
Notifications
You must be signed in to change notification settings - Fork 0
/
config_insal_spatial-dyn_rgcn_pytorch.yml
59 lines (48 loc) · 2.08 KB
/
config_insal_spatial-dyn_rgcn_pytorch.yml
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
runDescription: "DRB Inland Salinity RGCN prediction with spatial holdout - dynamic attributes"
# Input files
obs_file: "2_process/out/drb_SC_obs_PRMS.zarr"
attrs_file: "2_process/out/drb_attrs_PRMS.zarr"
dist_matrix_file: "1_fetch/out/drb_distance_matrix.npz"
# File containing the attribute names to use within the attrs_file
x_vars_file: "4_predict/out/dynamic_attrs.txt"
#river-dl code directory
code_dir: "../river-dl"
# Attribute names within obs_file
y_vars: ['mean_value']
#Attributs to take the log of for model training.
log_y_vars: False
# output location
out_dir: "4_predict/out/spatial/RGCN/dynamic_test1"
#random seed for training False==No seed, otherwise specify the seed
seed: 21023
lambdas: [1]
#Define number of epochs and the early stopping criteria (number of epochs)
epochs: 200
#setting this to False will turn off early stopping rounds
early_stopping: 10
#Hyperparameter settings
hidden_size: 20
dropout: 0.2
recurrent_dropout: 0.2
learning_rate: 0.005
trn_offset: 1.0
tst_val_offset: 1.0
seq_len: 365
# Define the spatial test segments (to be loaded from a file)
# Use 'None' if all segments may be used
# using f instead of file because file triggers preproc_utils to generate metadata that wouldn't exist for 'None'
# None for training means that val and test segs are removed to create the training data.
train_segs_f: 'None'
val_segs_f: '4_predict/out/spatial_val_reaches1.txt'
test_segs_f: '4_predict/out/spatial_test_reaches.txt'
#when True, the final file of predictions and observations are trimmed to the reaches in each of the 3 above files
spatial_write: True
# Define the earliest date, latest date, training, validation, and testing time periods
earliest_date: "4_predict/out/train_start.txt"
latest_date: "4_predict/out/test_end.txt"
train_start_date_f: "4_predict/out/train_start.txt"
train_end_date_f: "4_predict/out/test_end.txt"
val_start_date_f: "4_predict/out/train_start.txt"
val_end_date_f: "4_predict/out/test_end.txt"
test_start_date_f: "4_predict/out/train_start.txt"
test_end_date_f: "4_predict/out/test_end.txt"