Add params_to_tune for pf NNs (DeepARModel, TFTModel)

[Mr-Geekman]

🚀 Feature Request

Add params_to_tune for pf NNs (DeepARModel, TFTModel)

Proposal

Suggested grid for DeepARModel:

{
    "hidden_size": IntDistribution(4, 64, step=4),
    "rnn_layers": IntDistribution(4, 3, step=1),
    "dropout": Uniform(0, 0.5),
    "lr": LogUniformDistribution(1e-5, 1e-2),
}

Suggested grid for TFTModel:

{
    "hidden_size": IntDistribution(4, 64, step=4),
    "lstm_layers": IntDistribution(4, 3, step=1),
    "dropout": Uniform(0, 0.5),
    "attention_head_size": IntDistribution(2, 8, step=2),
    "lr": LogUniformDistribution(1e-5, 1e-2),
}

To discuss:

• Is tuning lr necessary?

Test cases

Look at #1184.

Additional context

No response

[martins0n]

The same comments for lr as for mlp rnn

[Mr-Geekman]

Closed by #1210.