def test_test_gages(self):
data_model = GagesModel.load_datamodel(
self.config_data.data_path["Temp"],
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow.npy',
forcing_file_name='test_forcing.npy',
attr_file_name='test_attr.npy',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(1):
pred, obs = master_test(data_model, epoch=self.test_epoch)
basin_area = data_model.data_source.read_attr(
data_model.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = data_model.data_source.read_attr(
data_model.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(data_model.data_source.data_config.data_path['Temp'],
self.test_epoch, pred, obs)
plot_we_need(data_model,
obs,
pred,
id_col="STAID",
lon_col="LNG_GAGE",
lat_col="LAT_GAGE")
def camels_lstm(args):
update_cfg(cfg, args)
random_seed = cfg.RANDOM_SEED
test_epoch = cfg.TEST_EPOCH
gpu_num = cfg.CTX
train_mode = cfg.TRAIN_MODE
cache = cfg.CACHE.STATE
print("train and test in CAMELS: \n")
config_data = GagesConfig(cfg)
camels531_gageid_file = os.path.join(config_data.data_path["DB"],
"camels531", "camels531.txt")
gauge_df = pd.read_csv(camels531_gageid_file, dtype={"GaugeID": str})
gauge_list = gauge_df["GaugeID"].values
all_sites_camels_531 = np.sort(
[str(gauge).zfill(8) for gauge in gauge_list])
gages_model = GagesModels(config_data,
screen_basin_area_huc4=False,
sites_id=all_sites_camels_531.tolist())
gages_model_train = gages_model.data_model_train
gages_model_test = gages_model.data_model_test
if cache:
save_datamodel(gages_model_train,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(gpu_num):
if train_mode:
master_train(gages_model_train, random_seed=random_seed)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
def test_forecast(self):
source_data = unserialize_pickle(self.data_source_test_file)
# 存储data_model，因为data_model里的数据如果直接序列化会比较慢，所以各部分分别序列化，dict的直接序列化为json文件，数据的HDF5
stat_dict = unserialize_json(self.stat_file)
data_flow = unserialize_numpy(self.flow_npy_file)
data_forcing = unserialize_numpy(self.forcing_npy_file)
data_attr = unserialize_numpy(self.attr_npy_file)
# dictFactorize.json is the explanation of value of categorical variables
var_dict = unserialize_json(self.var_dict_file)
f_dict = unserialize_json(self.f_dict_file)
t_s_dict = unserialize_json(self.t_s_dict_file)
data_model_test = DataModel(source_data, data_flow, data_forcing,
data_attr, var_dict, f_dict, stat_dict,
t_s_dict)
pred, obs = hydroDL.master_test(data_model_test)
print(pred)
print(obs)
serialize_numpy(pred, self.flow_pred_file)
serialize_numpy(obs, self.flow_obs_file)
def dor_lstm(args):
update_cfg(cfg, args)
random_seed = cfg.RANDOM_SEED
test_epoch = cfg.TEST_EPOCH
gpu_num = cfg.CTX
train_mode = cfg.TRAIN_MODE
dor = cfg.GAGES.attrScreenParams.DOR
cache = cfg.CACHE.STATE
print("train and test in some dor basins: \n")
config_data = GagesConfig(cfg)
gages_model = GagesModels(config_data, screen_basin_area_huc4=False, DOR=dor)
gages_model_train = gages_model.data_model_train
gages_model_test = gages_model.data_model_test
if cache:
save_datamodel(gages_model_train, data_source_file_name='data_source.txt',
stat_file_name='Statistics.json', flow_file_name='flow', forcing_file_name='forcing',
attr_file_name='attr', f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json', t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test, data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json', flow_file_name='test_flow',
forcing_file_name='test_forcing', attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json', var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(gpu_num):
if train_mode:
master_train(gages_model_train, random_seed=random_seed)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(gages_model_test.data_source.data_config.data_path['Temp'], test_epoch, pred, obs)
def pub_lstm(args):
update_cfg(cfg, args)
random_seed = cfg.RANDOM_SEED
test_epoch = cfg.TEST_EPOCH
gpu_num = cfg.CTX
train_mode = cfg.TRAIN_MODE
cache = cfg.CACHE.STATE
pub_plan = cfg.PUB_PLAN
plus = cfg.PLUS
dor = cfg.GAGES.attrScreenParams.DOR
split_num = cfg.SPLIT_NUM
print("train and test for PUB: \n")
config_data = GagesConfig(cfg)
if cache:
eco_names = [
("ECO2_CODE", 5.2), ("ECO2_CODE", 5.3), ("ECO2_CODE", 6.2),
("ECO2_CODE", 7.1), ("ECO2_CODE", 8.1), ("ECO2_CODE", 8.2),
("ECO2_CODE", 8.3), ("ECO2_CODE", 8.4), ("ECO2_CODE", 8.5),
("ECO2_CODE", 9.2), ("ECO2_CODE", 9.3), ("ECO2_CODE", 9.4),
("ECO2_CODE", 9.5), ("ECO2_CODE", 9.6), ("ECO2_CODE", 10.1),
("ECO2_CODE", 10.2), ("ECO2_CODE", 10.4), ("ECO2_CODE", 11.1),
("ECO2_CODE", 12.1), ("ECO2_CODE", 13.1)
]
quick_data_dir = os.path.join(config_data.data_path["DB"], "quickdata")
data_dir = os.path.join(quick_data_dir,
"conus-all_90-10_nan-0.0_00-1.0")
data_model_train = GagesModel.load_datamodel(
data_dir,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow.npy',
forcing_file_name='forcing.npy',
attr_file_name='attr.npy',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
data_model_test = GagesModel.load_datamodel(
data_dir,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow.npy',
forcing_file_name='test_forcing.npy',
attr_file_name='test_attr.npy',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
conus_sites_id = data_model_train.t_s_dict["sites_id"]
if pub_plan == 0:
"""do a pub test like freddy's"""
camels531_gageid_file = os.path.join(config_data.data_path["DB"],
"camels531", "camels531.txt")
gauge_df = pd.read_csv(camels531_gageid_file,
dtype={"GaugeID": str})
gauge_list = gauge_df["GaugeID"].values
all_sites_camels_531 = np.sort(
[str(gauge).zfill(8) for gauge in gauge_list])
sites_id_train = np.intersect1d(conus_sites_id,
all_sites_camels_531)
# basins not in CAMELS
sites_id_test = [
a_temp_site for a_temp_site in conus_sites_id
if a_temp_site not in all_sites_camels_531
]
assert (all(x < y
for x, y in zip(sites_id_test, sites_id_test[1:])))
elif pub_plan == 1 or pub_plan == 4:
source_data_dor1 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=-dor)
# basins with dams
source_data_withdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=[1, 100000])
# basins without dams
source_data_withoutdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=0)
sites_id_dor1 = source_data_dor1.all_configs['flow_screen_gage_id']
sites_id_withdams = source_data_withdams.all_configs[
'flow_screen_gage_id']
if pub_plan == 1:
sites_id_train = source_data_withoutdams.all_configs[
'flow_screen_gage_id']
sites_id_test = np.intersect1d(
np.array(sites_id_dor1),
np.array(sites_id_withdams)).tolist()
else:
sites_id_train = np.intersect1d(
np.array(sites_id_dor1),
np.array(sites_id_withdams)).tolist()
sites_id_test = source_data_withoutdams.all_configs[
'flow_screen_gage_id']
elif pub_plan == 2 or pub_plan == 5:
source_data_dor1 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=dor)
# basins without dams
source_data_withoutdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=0)
if pub_plan == 2:
sites_id_train = source_data_withoutdams.all_configs[
'flow_screen_gage_id']
sites_id_test = source_data_dor1.all_configs[
'flow_screen_gage_id']
else:
sites_id_train = source_data_dor1.all_configs[
'flow_screen_gage_id']
sites_id_test = source_data_withoutdams.all_configs[
'flow_screen_gage_id']
elif pub_plan == 3 or pub_plan == 6:
dor_1 = -dor
dor_2 = dor
source_data_dor1 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=dor_1)
# basins with dams
source_data_withdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=[1, 100000])
sites_id_dor1 = source_data_dor1.all_configs['flow_screen_gage_id']
sites_id_withdams = source_data_withdams.all_configs[
'flow_screen_gage_id']
source_data_dor2 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=dor_2)
if pub_plan == 3:
sites_id_train = np.intersect1d(
np.array(sites_id_dor1),
np.array(sites_id_withdams)).tolist()
sites_id_test = source_data_dor2.all_configs[
'flow_screen_gage_id']
else:
sites_id_train = source_data_dor2.all_configs[
'flow_screen_gage_id']
sites_id_test = np.intersect1d(
np.array(sites_id_dor1),
np.array(sites_id_withdams)).tolist()
else:
print("wrong plan")
sites_id_train = None
sites_id_test = None
train_sites_in_conus = np.intersect1d(conus_sites_id, sites_id_train)
test_sites_in_conus = np.intersect1d(conus_sites_id, sites_id_test)
if plus == 0:
all_index_lst_train_1 = []
# all sites come from train1 dataset
sites_lst_train = []
all_index_lst_test_1 = []
sites_lst_test_1 = []
all_index_lst_test_2 = []
sites_lst_test_2 = []
np.random.seed(random_seed)
kf = KFold(n_splits=split_num,
shuffle=True,
random_state=random_seed)
eco_name_chosen = []
for eco_name in eco_names:
eco_source_data = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
ecoregion=eco_name)
eco_sites_id = eco_source_data.all_configs[
'flow_screen_gage_id']
train_sites_id_inter = np.intersect1d(train_sites_in_conus,
eco_sites_id)
test_sites_id_inter = np.intersect1d(test_sites_in_conus,
eco_sites_id)
if train_sites_id_inter.size < split_num or test_sites_id_inter.size < 1:
continue
for train, test in kf.split(train_sites_id_inter):
all_index_lst_train_1.append(train)
sites_lst_train.append(train_sites_id_inter[train])
all_index_lst_test_1.append(test)
sites_lst_test_1.append(train_sites_id_inter[test])
if test_sites_id_inter.size < test.size:
all_index_lst_test_2.append(
np.arange(test_sites_id_inter.size))
sites_lst_test_2.append(test_sites_id_inter)
else:
test2_chosen_idx = np.random.choice(
test_sites_id_inter.size, test.size, replace=False)
all_index_lst_test_2.append(test2_chosen_idx)
sites_lst_test_2.append(
test_sites_id_inter[test2_chosen_idx])
eco_name_chosen.append(eco_name)
elif plus == -1:
print("camels pub, only do pub on the camels basins")
all_index_lst_train_1 = []
# all sites come from train1 dataset
sites_lst_train = []
all_index_lst_test_1 = []
sites_lst_test_1 = []
np.random.seed(random_seed)
kf = KFold(n_splits=split_num,
shuffle=True,
random_state=random_seed)
eco_name_chosen = []
for eco_name in eco_names:
eco_source_data = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
ecoregion=eco_name)
eco_sites_id = eco_source_data.all_configs[
'flow_screen_gage_id']
train_sites_id_inter = np.intersect1d(train_sites_in_conus,
eco_sites_id)
if train_sites_id_inter.size < split_num:
continue
for train, test in kf.split(train_sites_id_inter):
all_index_lst_train_1.append(train)
sites_lst_train.append(train_sites_id_inter[train])
all_index_lst_test_1.append(test)
sites_lst_test_1.append(train_sites_id_inter[test])
eco_name_chosen.append(eco_name)
elif plus == -2:
print(
"camels pub, only do pub on the camels basins, same with freddy's split method"
)
all_index_lst_train_1 = []
# all sites come from train1 dataset
sites_lst_train = []
all_index_lst_test_1 = []
sites_lst_test_1 = []
np.random.seed(random_seed)
kf = KFold(n_splits=split_num,
shuffle=True,
random_state=random_seed)
for train, test in kf.split(train_sites_in_conus):
all_index_lst_train_1.append(train)
sites_lst_train.append(train_sites_in_conus[train])
all_index_lst_test_1.append(test)
sites_lst_test_1.append(train_sites_in_conus[test])
else:
sites_lst_train = []
sites_lst_test_1 = []
sites_lst_test_2 = []
np.random.seed(random_seed)
kf = KFold(n_splits=split_num,
shuffle=True,
random_state=random_seed)
eco_name_chosen = []
for eco_name in eco_names:
eco_source_data = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
ecoregion=eco_name)
eco_sites_id = eco_source_data.all_configs[
'flow_screen_gage_id']
sites_id_inter_1 = np.intersect1d(train_sites_in_conus,
eco_sites_id)
sites_id_inter_2 = np.intersect1d(test_sites_in_conus,
eco_sites_id)
if sites_id_inter_1.size < sites_id_inter_2.size:
if sites_id_inter_1.size < split_num:
continue
for train, test in kf.split(sites_id_inter_1):
sites_lst_train_1 = sites_id_inter_1[train]
sites_lst_test_1.append(sites_id_inter_1[test])
chosen_lst_2 = random_choice_no_return(
sites_id_inter_2, [train.size, test.size])
sites_lst_train_2 = chosen_lst_2[0]
sites_lst_test_2.append(chosen_lst_2[1])
sites_lst_train.append(
np.sort(
np.append(sites_lst_train_1,
sites_lst_train_2)))
else:
if sites_id_inter_2.size < split_num:
continue
for train, test in kf.split(sites_id_inter_2):
sites_lst_train_2 = sites_id_inter_2[train]
sites_lst_test_2.append(sites_id_inter_2[test])
chosen_lst_1 = random_choice_no_return(
sites_id_inter_1, [train.size, test.size])
sites_lst_train_1 = chosen_lst_1[0]
sites_lst_test_1.append(chosen_lst_1[1])
sites_lst_train.append(
np.sort(
np.append(sites_lst_train_1,
sites_lst_train_2)))
eco_name_chosen.append(eco_name)
for i in range(split_num):
sites_ids_train_ilst = [
sites_lst_train[j] for j in range(len(sites_lst_train))
if j % split_num == i
]
sites_ids_train_i = np.sort(
reduce(lambda x, y: np.hstack((x, y)), sites_ids_train_ilst))
sites_ids_test_ilst_1 = [
sites_lst_test_1[j] for j in range(len(sites_lst_test_1))
if j % split_num == i
]
sites_ids_test_i_1 = np.sort(
reduce(lambda x, y: np.hstack((x, y)), sites_ids_test_ilst_1))
if plus >= 0:
sites_ids_test_ilst_2 = [
sites_lst_test_2[j] for j in range(len(sites_lst_test_2))
if j % split_num == i
]
sites_ids_test_i_2 = np.sort(
reduce(lambda x, y: np.hstack((x, y)),
sites_ids_test_ilst_2))
config_data_i = GagesConfig.set_subdir(cfg, str(i))
gages_model_train_i = GagesModel.update_data_model(
config_data_i,
data_model_train,
sites_id_update=sites_ids_train_i,
data_attr_update=True,
screen_basin_area_huc4=False)
gages_model_test_baseline_i = GagesModel.update_data_model(
config_data_i,
data_model_test,
sites_id_update=sites_ids_train_i,
data_attr_update=True,
train_stat_dict=gages_model_train_i.stat_dict,
screen_basin_area_huc4=False)
gages_model_test_i_1 = GagesModel.update_data_model(
config_data_i,
data_model_test,
sites_id_update=sites_ids_test_i_1,
data_attr_update=True,
train_stat_dict=gages_model_train_i.stat_dict,
screen_basin_area_huc4=False)
if plus >= 0:
gages_model_test_i_2 = GagesModel.update_data_model(
config_data_i,
data_model_test,
sites_id_update=sites_ids_test_i_2,
data_attr_update=True,
train_stat_dict=gages_model_train_i.stat_dict,
screen_basin_area_huc4=False)
save_datamodel(gages_model_train_i,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test_baseline_i,
data_source_file_name='test_data_source_base.txt',
stat_file_name='test_Statistics_base.json',
flow_file_name='test_flow_base',
forcing_file_name='test_forcing_base',
attr_file_name='test_attr_base',
f_dict_file_name='test_dictFactorize_base.json',
var_dict_file_name='test_dictAttribute_base.json',
t_s_dict_file_name='test_dictTimeSpace_base.json')
save_datamodel(gages_model_test_i_1,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
if plus >= 0:
save_datamodel(gages_model_test_i_2,
data_source_file_name='test_data_source_2.txt',
stat_file_name='test_Statistics_2.json',
flow_file_name='test_flow_2',
forcing_file_name='test_forcing_2',
attr_file_name='test_attr_2',
f_dict_file_name='test_dictFactorize_2.json',
var_dict_file_name='test_dictAttribute_2.json',
t_s_dict_file_name='test_dictTimeSpace_2.json')
print("save ecoregion " + str(i) + " data model")
with torch.cuda.device(gpu_num):
if train_mode:
for i in range(split_num):
data_model = GagesModel.load_datamodel(
config_data.data_path["Temp"],
str(i),
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow.npy',
forcing_file_name='forcing.npy',
attr_file_name='attr.npy',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
master_train(data_model, random_seed=random_seed)
for i in range(split_num):
data_model_baseline = GagesModel.load_datamodel(
config_data.data_path["Temp"],
str(i),
data_source_file_name='test_data_source_base.txt',
stat_file_name='test_Statistics_base.json',
flow_file_name='test_flow_base.npy',
forcing_file_name='test_forcing_base.npy',
attr_file_name='test_attr_base.npy',
f_dict_file_name='test_dictFactorize_base.json',
var_dict_file_name='test_dictAttribute_base.json',
t_s_dict_file_name='test_dictTimeSpace_base.json')
data_model = GagesModel.load_datamodel(
config_data.data_path["Temp"],
str(i),
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow.npy',
forcing_file_name='test_forcing.npy',
attr_file_name='test_attr.npy',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
if plus >= 0:
data_model_2 = GagesModel.load_datamodel(
config_data.data_path["Temp"],
str(i),
data_source_file_name='test_data_source_2.txt',
stat_file_name='test_Statistics_2.json',
flow_file_name='test_flow_2.npy',
forcing_file_name='test_forcing_2.npy',
attr_file_name='test_attr_2.npy',
f_dict_file_name='test_dictFactorize_2.json',
var_dict_file_name='test_dictAttribute_2.json',
t_s_dict_file_name='test_dictTimeSpace_2.json')
pred_baseline, obs_baseline = master_test(data_model_baseline,
epoch=test_epoch,
save_file_suffix="base")
basin_area_baseline = data_model_baseline.data_source.read_attr(
data_model_baseline.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep_baseline = data_model_baseline.data_source.read_attr(
data_model_baseline.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep_baseline = mean_prep_baseline / 365 * 10
pred_baseline = _basin_norm(pred_baseline,
basin_area_baseline,
mean_prep_baseline,
to_norm=False)
obs_baseline = _basin_norm(obs_baseline,
basin_area_baseline,
mean_prep_baseline,
to_norm=False)
save_result(
data_model_baseline.data_source.data_config.data_path['Temp'],
test_epoch,
pred_baseline,
obs_baseline,
pred_name='flow_pred_base',
obs_name='flow_obs_base')
pred, obs = master_test(data_model, epoch=test_epoch)
basin_area = data_model.data_source.read_attr(
data_model.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = data_model.data_source.read_attr(
data_model.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(data_model.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
if plus >= 0:
pred_2, obs_2 = master_test(data_model_2,
epoch=test_epoch,
save_file_suffix="2")
basin_area_2 = data_model_2.data_source.read_attr(
data_model_2.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep_2 = data_model_2.data_source.read_attr(
data_model_2.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep_2 = mean_prep_2 / 365 * 10
pred_2 = _basin_norm(pred_2,
basin_area_2,
mean_prep_2,
to_norm=False)
obs_2 = _basin_norm(obs_2,
basin_area_2,
mean_prep_2,
to_norm=False)
save_result(
data_model_2.data_source.data_config.data_path['Temp'],
test_epoch,
pred_2,
obs_2,
pred_name='flow_pred_2',
obs_name='flow_obs_2')
def test_some_reservoirs(self):
print("train and test in basins with different combination: \n")
dam_plan = self.dam_plan
config_data = self.config_data
test_epoch = self.test_epoch
if dam_plan == 2:
dam_num = 0
dor = self.config_file.GAGES.attrScreenParams.DOR
source_data_dor1 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=dor)
# basins with dams
source_data_withoutdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=dam_num)
sites_id_dor1 = source_data_dor1.all_configs['flow_screen_gage_id']
sites_id_withoutdams = source_data_withoutdams.all_configs[
'flow_screen_gage_id']
sites_id_chosen = np.sort(
np.union1d(np.array(sites_id_dor1),
np.array(sites_id_withoutdams))).tolist()
elif dam_plan == 3:
dam_num = [1, 100000]
# basins with dams
source_data_withdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=dam_num)
sites_id_chosen = source_data_withdams.all_configs[
'flow_screen_gage_id']
else:
print("wrong choice")
sites_id_chosen = None
gages_model = GagesModels(config_data,
screen_basin_area_huc4=False,
sites_id=sites_id_chosen)
gages_model_train = gages_model.data_model_train
gages_model_test = gages_model.data_model_test
if self.cache:
save_datamodel(gages_model_train,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(self.gpu_num):
if self.train_mode:
master_train(gages_model_train, random_seed=self.random_seed)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(
gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
def dam_lstm(args):
update_cfg(cfg, args)
random_seed = cfg.RANDOM_SEED
test_epoch = cfg.TEST_EPOCH
gpu_num = cfg.CTX
train_mode = cfg.TRAIN_MODE
dor = cfg.GAGES.attrScreenParams.DOR
cache = cfg.CACHE.STATE
print("train and test in basins with dams: \n")
config_data = GagesConfig(cfg)
source_data_dor1 = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
DOR=dor)
# basins with dams
source_data_withdams = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
dam_num=[1, 100000])
sites_id_dor1 = source_data_dor1.all_configs['flow_screen_gage_id']
sites_id_withdams = source_data_withdams.all_configs['flow_screen_gage_id']
sites_id_chosen = np.intersect1d(np.array(sites_id_dor1),
np.array(sites_id_withdams)).tolist()
gages_model = GagesModels(config_data,
screen_basin_area_huc4=False,
sites_id=sites_id_chosen)
gages_model_train = gages_model.data_model_train
gages_model_test = gages_model.data_model_test
if cache:
save_datamodel(gages_model_train,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(gpu_num):
if train_mode:
master_train(gages_model_train, random_seed=random_seed)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
# train_stat_dict=gages_model_train.stat_dict,
# screen_basin_area_huc4=False)
gages_model_train = GagesModel.update_data_model(
config_data,
data_model_train,
data_attr_update=True,
screen_basin_area_huc4=False)
gages_model_test = GagesModel.update_data_model(
config_data,
data_model_test,
data_attr_update=True,
train_stat_dict=gages_model_train.stat_dict,
screen_basin_area_huc4=False)
with torch.cuda.device(gpu_lst[i]):
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(
gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
plot_we_need(gages_model_test,
obs,
pred,
def synergy_ecoregion(args):
update_cfg(cfg, args)
cache = cfg.CACHE.STATE
train_mode = cfg.TRAIN_MODE
test_epoch = cfg.TEST_EPOCH
config_data = GagesConfig(cfg)
eco_names = [("ECO2_CODE", 5.2), ("ECO2_CODE", 5.3), ("ECO2_CODE", 6.2),
("ECO2_CODE", 7.1), ("ECO2_CODE", 8.1), ("ECO2_CODE", 8.2),
("ECO2_CODE", 8.3), ("ECO2_CODE", 8.4), ("ECO2_CODE", 8.5),
("ECO2_CODE", 9.2), ("ECO2_CODE", 9.3), ("ECO2_CODE", 9.4),
("ECO2_CODE", 9.5), ("ECO2_CODE", 9.6), ("ECO2_CODE", 10.1),
("ECO2_CODE", 10.2), ("ECO2_CODE", 10.4), ("ECO2_CODE", 11.1),
("ECO2_CODE", 12.1), ("ECO2_CODE", 13.1)]
quick_data_dir = os.path.join(config_data.data_path["DB"], "quickdata")
data_dir = os.path.join(quick_data_dir, "conus-all_90-10_nan-0.0_00-1.0")
data_model_train = GagesModel.load_datamodel(
data_dir,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow.npy',
forcing_file_name='forcing.npy',
attr_file_name='attr.npy',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
data_model_test = GagesModel.load_datamodel(
data_dir,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow.npy',
forcing_file_name='test_forcing.npy',
attr_file_name='test_attr.npy',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
for eco_name in eco_names:
source_data = GagesSource.choose_some_basins(
config_data,
config_data.model_dict["data"]["tRangeTrain"],
screen_basin_area_huc4=False,
ecoregion=eco_name)
sites_id = source_data.all_configs['flow_screen_gage_id']
sites_id_inter = np.intersect1d(data_model_train.t_s_dict["sites_id"],
sites_id)
if sites_id_inter.size < 1:
continue
config_data = GagesConfig.set_subdir(cfg, str(eco_name[1]))
gages_model_train = GagesModel.update_data_model(
config_data,
data_model_train,
sites_id_update=sites_id,
data_attr_update=True,
screen_basin_area_huc4=False)
gages_model_test = GagesModel.update_data_model(
config_data,
data_model_test,
sites_id_update=sites_id,
data_attr_update=True,
train_stat_dict=gages_model_train.stat_dict,
screen_basin_area_huc4=False)
if cache:
save_datamodel(gages_model_train,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
print("save ecoregion " + str(eco_name[1]) + " data model")
with torch.cuda.device(0):
if train_mode:
master_train(gages_model_train)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(
gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)
def conus_lstm(args):
update_cfg(cfg, args)
random_seed = cfg.RANDOM_SEED
test_epoch = cfg.TEST_EPOCH
gpu_num = cfg.CTX
train_mode = cfg.TRAIN_MODE
print("train and test in CONUS: \n")
print(cfg)
config_data = GagesConfig(cfg)
gages_model = GagesModels(config_data, screen_basin_area_huc4=False)
gages_model_train = gages_model.data_model_train
gages_model_test = gages_model.data_model_test
if cfg.CACHE.GEN_QUICK_DATA:
if not os.path.isdir(cfg.CACHE.DATA_DIR):
os.makedirs(cfg.CACHE.DATA_DIR)
save_quick_data(gages_model_train,
cfg.CACHE.DATA_DIR,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_quick_data(gages_model_test,
cfg.CACHE.DATA_DIR,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
if cfg.CACHE.STATE:
save_datamodel(gages_model_train,
data_source_file_name='data_source.txt',
stat_file_name='Statistics.json',
flow_file_name='flow',
forcing_file_name='forcing',
attr_file_name='attr',
f_dict_file_name='dictFactorize.json',
var_dict_file_name='dictAttribute.json',
t_s_dict_file_name='dictTimeSpace.json')
save_datamodel(gages_model_test,
data_source_file_name='test_data_source.txt',
stat_file_name='test_Statistics.json',
flow_file_name='test_flow',
forcing_file_name='test_forcing',
attr_file_name='test_attr',
f_dict_file_name='test_dictFactorize.json',
var_dict_file_name='test_dictAttribute.json',
t_s_dict_file_name='test_dictTimeSpace.json')
with torch.cuda.device(gpu_num):
if train_mode:
master_train(gages_model_train, random_seed=random_seed)
pred, obs = master_test(gages_model_test, epoch=test_epoch)
basin_area = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['DRAIN_SQKM'],
is_return_dict=False)
mean_prep = gages_model_test.data_source.read_attr(
gages_model_test.t_s_dict["sites_id"], ['PPTAVG_BASIN'],
is_return_dict=False)
mean_prep = mean_prep / 365 * 10
pred = _basin_norm(pred, basin_area, mean_prep, to_norm=False)
obs = _basin_norm(obs, basin_area, mean_prep, to_norm=False)
save_result(gages_model_test.data_source.data_config.data_path['Temp'],
test_epoch, pred, obs)