def main():
train_dataloader, validation_dataloader = get_dataloaders()
model = Model()
logger = NeptuneLogger(project='OpenClimateFix/predict-pv-yield')
logger.log_hyperparams(params)
_LOG.info(f'logger.version = {logger.version}')
trainer = pl.Trainer(gpus=0, max_epochs=1, logger=logger)
trainer.fit(model, train_dataloader)
def main():
train_dataloader, validation_dataloader = get_dataloaders(n_validation_data=1000, cloud='aws')
model = Model()
# logger = NeptuneLogger(project="OpenClimateFix/predict-pv-yield")
# _LOG.info(f"logger.version = {logger.version}")
trainer = pl.Trainer(gpus=0, max_epochs=10)
# dont need to train baseline model
# trainer.fit(model, train_dataloaders=train_dataloader, val_dataloaders=validation_dataloader)
trainer.validate(model, validation_dataloader)
def main():
train_dataloader, validation_dataloader = get_dataloaders(
n_train_data=10, n_validation_data=10)
model = Model()
logger = NeptuneLogger(project='OpenClimateFix/predict-pv-yield')
logger.log_hyperparams(model_configuration_default)
_LOG.info(f'logger.version = {logger.version}')
trainer = pl.Trainer(gpus=1, max_epochs=10, logger=logger)
trainer.fit(model,
train_dataloaders=train_dataloader,
val_dataloaders=validation_dataloader)
# run validation
trainer.validate(model, validation_dataloader)
def main():
train_dataloader, validation_dataloader = get_dataloaders(
n_train_data=24900,
n_validation_data=1000,
data_path="gs://solar-pv-nowcasting-data/prepared_ML_training_data/v4/",
cloud="gcp",
)
model = Model()
logger = NeptuneLogger(project="OpenClimateFix/predict-pv-yield")
logger.log_hyperparams(model_configuration_default)
_LOG.info(f"logger.version = {logger.version}")
trainer = pl.Trainer(gpus=1, max_epochs=10, logger=logger)
trainer.fit(model,
train_dataloaders=train_dataloader,
val_dataloaders=validation_dataloader)
# run validation
trainer.validate(model, validation_dataloader)
from predict_pv_yield.models.conv3d.model import Model
from predict_pv_yield.data.dataloader import get_dataloaders
from pytorch_lightning.utilities.cloud_io import load as pl_load
import torch
weights = './weights/conv3d/last.ckpt'
checkpoint = pl_load(weights, map_location=torch.device('cpu'))
model = Model(conv3d_channels=32,
fc1_output_features=32,
fc2_output_features=16,
fc3_output_features=16,
include_time=False,
number_of_conv3d_layers=4)
model.load_from_checkpoint(weights)
train_dataset, validation_dataset = get_dataloaders()
checkpoint = pl_load(weights, map_location=torch.device("cpu"))
model = Model(
conv3d_channels=32,
fc1_output_features=128,
fc2_output_features=128,
fc3_output_features=64,
include_time=True,
forecast_len=12,
history_len=6,
number_of_conv3d_layers=6,
)
model.load_state_dict(checkpoint["state_dict"])
train_dataset, validation_dataset = get_dataloaders(
cloud="gcp",
data_path="gs://solar-pv-nowcasting-data/prepared_ML_training_data/v4/")
validation_dataset = iter(validation_dataset)
x = next(validation_dataset)
y_hat_all = model(x)
# plot one
batch_index = 0
y = x["pv_yield"][batch_index][7:, 0].detach().numpy()
y_hat = y_hat_all[batch_index].detach().numpy()
time = pd.to_datetime(
x["sat_datetime_index"][batch_index][7:].detach().numpy(), unit="s")
fig = plot_one_result(x=time, y=y, y_hat=y_hat)
fig.show(renderer="browser")