if params['net_type'] == 'embeddings': model = TabNet(params)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# TODO: move problem type to params
looper = Looper(device, 'cls', add_metrics)
model.cuda()
torch.save(model.state_dict(),EXP_NAME+'.pth')
optimizer = optim.AdamW(model.parameters(), lr=params['lr'], weight_decay=1e-06)
loss_fn = torch.nn.CrossEntropyLoss()
val_funcs = [roc_auc_score, accuracy_score]
print(f'Training neural {params["net_type"]}')
looper.train(model,
10,
train_set,
val_set,
optimizer,
loss_fn,
val_funcs)
torch.save(model.state_dict(),EXP_NAME+'.pth')
return float(max(looper.history()['roc_auc_score']))
if __name__ == '__main__':
run = ex._create_run()
run(run.config)
