def test_DataLoader(self):
smb = Glacier_dmdt("JAKOBSHAVN_ISBRAE",
1980,
2002,
path="../glaicer_dmdt.csv")
data = ERA5Datasets("JAKOBSHAVN_ISBRAE",
1980,
2002,
path="../ECMWF_reanalysis_data")
dataset = GlacierDataset([data, data], [smb, smb])
loader = DataLoader(dataset, batch_size=1)
for feature, s in loader:
self.assertEqual(feature.shape, (1, 5, 289, 12))
self.assertEqual(s.shape, (1, ))
def setUpClass(cls):
smb = Glacier_dmdt("JAKOBSHAVN_ISBRAE",
1980,
2002,
path="../glaicer_dmdt.csv")
data = ERA5Datasets("JAKOBSHAVN_ISBRAE",
1980,
2002,
path="../ECMWF_reanalysis_data")
dataset = GlacierDataset([data], [smb])
cls.loader = DataLoader(dataset, batch_size=16)
cls.device = torch.device(
"cuda" if torch.cuda.is_available() else "cpu")
cls.train_loader = cls.loader
cls.test_loader = cls.loader
def test_DMDTLoader(self):
smb = Glacier_dmdt("JAKOBSHAVN_ISBRAE",
1980,
2002,
path="../glaicer_dmdt.csv")
self.assertEqual(len(smb), 2002 - 1980)
show=False,
device=cuda, epochs=150, lr=0.002, reg=0.001, save_every=20, eval_every=1, test_split_at=split_at,
critic=loss_function, optimizer=torch.optim.Adam, save_path="saved_models")
glaicer_selected = pd.read_csv("Glacier_select.csv")
glaciers = list(glaicer_selected["NAME"])
end_year = 2018
for name in glaciers:
cond_1 = glaicer_selected["NAME"] == name
years = int(list(glaicer_selected[cond_1]["Years"])[0])
cal_start_year = lambda years: (2018 - years) if ((2018 - years) > 1979) else 1979
start_year = cal_start_year(years)
test_years = int(years * 0.2)
train_smb = Glacier_dmdt(name, start_year, end_year - test_years, path="glacier_dmdt.csv")
train_data = ERA5Datasets(name, start_year, end_year - test_years, path="ECMWF_reanalysis_data")
train_dataset = GlacierDataset([train_data], [train_smb])
train_loader = DataLoader(train_dataset, batch_size=1, shuffle=False)
second = train_data[1].shape[1]
if second <48:
continue
test_smb = Glacier_dmdt(name, end_year - test_years, end_year, path="glacier_dmdt.csv")
test_data = ERA5Datasets(name, end_year - test_years, end_year, path="ECMWF_reanalysis_data")
test_dataset = GlacierDataset([test_data], [test_smb])
test_loader = DataLoader(test_dataset, batch_size=1, shuffle=False)
train(name, train_loader, test_data, test_smb,int(years * 0.8))
# print(glaciers)
for name in glaciers:
try:
new_df = glacier_info[glacier_info["NAME"] == name]
start_year = 2018 - int(new_df["Years"])
if start_year < 1980:
start_year = 1980
last_year_smb = True
dataset = GlacierDataset3D(name,
start_year,
2018,
last_year=last_year_smb,
path="glacier_dmdt.csv")
# dataset = GlacierDatasetNoPadding3D(name, start_year, 2018, last_year=last_year_smb, path="glacier_dmdt.csv")
datasets = train_val_dataset(dataset, val_split=0.3, shuffle=False)
last_year_dmdt = Glacier_dmdt(name,
start_year - 1,
2017,
path="glacier_dmdt.csv")
test_last_year_dmdt = train_val_dataset(last_year_dmdt,
val_split=0.3)['val']
train_loader = DataLoader(datasets['train'], batch_size=1)
test_dataset = datasets['val']
extractor = SeparateFeatureExtractor3D(output_dim=256,
layers=[
TWCNN2D(),
TWCNN2D(),
TWCNN2D(),
TWCNN2D(),
TWCNN2D(),
TWCNN2D(),
TWCNN2D()
])
if __name__ == '__main__':
# set dataset
glacier_name = "STORSTROMMEN"
start, mid, end = 1980, 2008, 2018
last_year_smb = True
hidden_dim = 256
reg = 0
save_every = 1
eval_every = 1
epochs = 50
run_hcnn = True
from datasets import Glacier_dmdt, ERA5Datasets, GlacierDataset
from models import GlacierModel, LSTMPredictor, HCNN, VCNN, SeparateFeatureExtractor
train_smb = Glacier_dmdt(glacier_name, start, mid, path="glacier_dmdt.csv")
train_data = ERA5Datasets(glacier_name,
start,
mid,
path="ECMWF_reanalysis_data")
test_smb = Glacier_dmdt(glacier_name, start, end, path="glacier_dmdt.csv")
test_last_year_dmdt = Glacier_dmdt(glacier_name,
start - 1,
end - 1,
path="glacier_dmdt.csv")
test_data = ERA5Datasets(glacier_name,
start,
end,
path="ECMWF_reanalysis_data")
glacier_dataset = GlacierDataset([train_data], [train_smb],
last_year=last_year_smb)
import os
import torch
import argparse
from datasets import ERA5Datasets, Glacier_dmdt
parser = argparse.ArgumentParser(description='Glacier SMB prediction')
parser.add_argument('-model', type=str, help='Path to the saved_model')
parser.add_argument('-data', type=str, help='Path to the dataset', default="ECMWF_reanalysis_data")
parser.add_argument('-glacier', type=str, help='Glacier name')
parser.add_argument('-year', type=int, help='Glacier name')
args = parser.parse_args()
if args.year < 1980 or args.year > 2018:
print("[ERROR] Year not within the available range")
if not os.path.exists(args.model):
print("[ERROR] Model file not exist")
model = torch.load(args.model)
dataset = ERA5Datasets(args.glacier, 1980, 2018, path=args.data)
smb = Glacier_dmdt(args.glacier, 1980, 2018, path="glacier_dmdt.csv")
result = model(dataset[args.year - 1980].unsqueeze(0))
print("[INFO] Predicted SMB for {} in {} is: {}".format(args.glacier, args.year, result.item()))
print("[INFO] Actual SMB for {} in {} is: {}".format(args.glacier, args.year, smb[args.year - 1980]))