optimize_hyperparameters() uses QuantileLoss()

[Madr75]

• PyTorch-Forecasting version: 0.9.2
• PyTorch version: 1.10.1
• Python version: 3.8.12
• Operating System: win10 x64

Expected behavior

I executed code with optimize_hyperparameters function to find optimal parameters of a TFT network for a multiclass classification task using CrossEntropy() loss function: tft = TemporalFusionTransformer.from_dataset( training, learning_rate=0.03, lstm_layers = 2, hidden_size=64, attention_head_size=4, dropout=0.15, hidden_continuous_size=48,

output_size=7,

        **loss=CrossEntropy(),**
        log_interval=10,  # uncomment for learning rate finder and otherwise, e.g. to 10 for logging every 10 batches
        reduce_on_plateau_patience=4,
    )

study = optimize_hyperparameters( train_dataloader, val_dataloader, model_path="optuna_test", n_trials=200, max_epochs=20, gradient_clip_val_range=(0.01, 1.0), hidden_size_range=(16, 96), hidden_continuous_size_range=(16, 64), attention_head_size_range=(1, 4), learning_rate_range=(0.001, 0.1), dropout_range=(0.1, 0.3), trainer_kwargs=dict(limit_train_batches=20), reduce_on_plateau_patience=4, use_learning_rate_finder=False, # use Optuna to find ideal learning rate or use in-built learning rate finder )

Actual behavior

However, result was optimize_hyperparameters() function uses QuantileLoss() but I expected CrossEntropy() I see it in TensorBoard. Any guess how to pass other loss function.

[misterspook]

You can edit the tuning.py script and change it to your preferred loss function.