config_insal_temporal-minstatdyn_rgcn_pytorch.yml

runDescription: "DRB Inland Salinity RGCN prediction with temporal holdout - min static and 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/min_static_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/temporal/RGCN/min_static_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'
train_segs_f: 'None'
val_segs_f: 'None'
test_segs_f: 'None'
#when True, the final file of predictions and observations are trimmed to the reaches in each of the 3 above files
spatial_write: False

# 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/val_end4.txt"
val_start_date_f: "4_predict/out/val_start5.txt"
val_end_date_f: "4_predict/out/train_end.txt"
test_start_date_f: "4_predict/out/test_start.txt"
test_end_date_f: "4_predict/out/test_end.txt"