Ray error with AutoHINT

[candalfigomoro]

I'm trying to use AutoHINT in this way:

model = AutoHINT(
    NHITS,
    h=horizon,
    loss=GMM(n_components=10, quantiles=quantiles),
    valid_loss=sCRPS(quantiles=quantiles),
    S=S_df.values,
    config=config,
    search_alg=HyperOptSearch(),
    num_samples=2,
    verbose=True,
)

But I get this error:

ValueError: The actor ImplicitFunc is too large (235 MiB > FUNCTION_SIZE_ERROR_THRESHOLD=95 MiB). Check that its definition is not implicitly capturing a large array or other object in scope. Tip: use ray.put() to put large objects in the Ray object store.

The same config (without the "reconciliation" parameter) works if I use AutoNHITS instead of AutoHINT.

[kdgutier]

Hey @candalfigomoro,

Thanks for checking the AutoHINT. We are still on early stages of its development.

There are two AutoHINT versions: https://github.com/Nixtla/neuralforecast/blob/main/nbs/models.ipynb

• In the first version, which I suggest for the moment, I fit an AutoNHITS that goes into HINT.
• In the second version (that needs debugging) I attempt to do the whole hyperparameter estimation of the HINT.