Closed Netanelshoshan closed 3 months ago
So far it's performing better than expected and way faster than the tf version. (backtested on 2 pairs only) will need to run more tests and dryrun!
// Trying hard not to overfit. The model is fairly strict and doesn't generate a lot of signals. "model_training_parameters": { "learning_rate": 3e-3, "trainer_kwargs": { "n_steps": null, "batch_size": 32, "n_epochs": 10 }, "model_kwargs": { "hidden_dim": 128, "dropout_percent": 0.4, "n_layer": 3, "window_size": 5 } }