def save_to_csv(PyTorchExperimentLogger, data, path, header=None):
"""
Saves a numpy array to csv in the experiment save dir
Args:
data: The array to be stored as a save file
path: sub path in the save folder (or simply filename)
"""
folder_path = create_folder(PyTorchExperimentLogger.save_dir,
os.path.dirname(path))
file_path = folder_path + '/' + os.path.basename(path)
if not file_path.endswith('.csv'):
file_path += '.csv'
np.savetxt(file_path, data, delimiter=',', header=header, comments='')
return
from eICU_preprocessing.split_train_test import create_folder
from models.run_tpc import TPC
from models.initialise_arguments import initialise_tpc_arguments
if __name__ == '__main__':
c = initialise_tpc_arguments()
c['mode'] = 'test'
c['exp_name'] = 'TempWeightShare'
c['model_type'] = 'temp_only'
c['share_weights'] = True
c['temp_kernels'] = [32] * c['n_layers']
log_folder_path = create_folder('models/experiments/final', c.exp_name)
temp_weight_share = TPC(
config=c,
n_epochs=c.n_epochs,
name=c.exp_name,
base_dir=log_folder_path,
explogger_kwargs={'folder_format': '%Y-%m-%d_%H%M%S{run_number}'})
temp_weight_share.run()
from eICU_preprocessing.split_train_test import create_folder
from models.run_lstm import BaselineLSTM
from models.initialise_arguments import initialise_lstm_arguments
from models.final_experiment_scripts.best_hyperparameters import best_lstm
if __name__ == '__main__':
c = initialise_lstm_arguments()
c['exp_name'] = 'StandardLSTM'
c['dataset'] = 'MIMIC'
c['task'] = 'mortality'
c = best_lstm(c)
log_folder_path = create_folder('models/experiments/final/MIMIC/mortality',
c.exp_name)
baseline_lstm = BaselineLSTM(
config=c,
n_epochs=c.n_epochs,
name=c.exp_name,
base_dir=log_folder_path,
explogger_kwargs={'folder_format': '%Y-%m-%d_%H%M%S{run_number}'})
baseline_lstm.run()
c['n_layers'] = random.choice(param_grid['n_layers'])
c['kernel_size'] = random.choice(param_grid['kernel_size'][c['n_layers']])
c['temp_kernels'] = [random.choice(param_grid['temp_kernels'])
] * c['n_layers']
c['point_sizes'] = [random.choice(param_grid['point_sizes'])
] * c['n_layers']
c['learning_rate'] = round(random.choice(param_grid['learning_rate']), 5)
c['batch_size'] = random.choice(param_grid['batch_size'])
c['temp_dropout_rate'] = random.choice(param_grid['temp_dropout_rate'])
return c
if __name__ == '__main__':
for i in range(50):
try:
c = get_hyperparam_config('eICU')
log_folder_path = create_folder(
'models/experiments/hyperparameters/eICU', c.exp_name)
tpc = TPC(config=c,
n_epochs=c.n_epochs,
name=c.exp_name,
base_dir=log_folder_path,
explogger_kwargs={
'folder_format': '%Y-%m-%d_%H%M%S{run_number}'
})
tpc.run()
except RuntimeError:
continue
from eICU_preprocessing.split_train_test import create_folder
from models.run_lstm import BaselineLSTM
from models.initialise_arguments import initialise_lstm_arguments
from models.final_experiment_scripts.best_hyperparameters import best_cw_lstm
if __name__ == '__main__':
c = initialise_lstm_arguments()
c['exp_name'] = 'ChannelwiseLSTM'
c['dataset'] = 'MIMIC'
c['task'] = 'multitask'
c = best_cw_lstm(c)
log_folder_path = create_folder('models/experiments/final/MIMIC/multitask',
c.exp_name)
channelwise_lstm = BaselineLSTM(
config=c,
n_epochs=c.n_epochs,
name=c.exp_name,
base_dir=log_folder_path,
explogger_kwargs={'folder_format': '%Y-%m-%d_%H%M%S{run_number}'})
channelwise_lstm.run()
def setup(self):
self.setup_template()
self.model = TempPointConv(
config=self.config,
F=self.train_datareader.F,
D=self.train_datareader.D,
no_flat_features=self.train_datareader.no_flat_features).to(
device=self.device)
self.elog.print(self.model)
self.optimiser = Adam(self.model.parameters(),
lr=self.config.learning_rate,
weight_decay=self.config.L2_regularisation)
return
if __name__ == '__main__':
c = initialise_tpc_arguments()
c['exp_name'] = 'TPC'
log_folder_path = create_folder(
'models/experiments/{}/{}'.format(c.dataset, c.task), c.exp_name)
tpc = TPC(
config=c,
n_epochs=c.n_epochs,
name=c.exp_name,
base_dir=log_folder_path,
explogger_kwargs={'folder_format': '%Y-%m-%d_%H%M%S{run_number}'})
tpc.run()
