Deprecate signature as argument

[blythed]

This PR disables the need to add signature='*args,**kwargs' or similar to a model.

[jieguangzhou]

I thought about it and I’m not entirely sure if the concept of signature is really necessary for the model.

For an individual model, it actually doesn’t seem to need a signature.

The signature only comes into play when we use a trainer or a listener.

In fact, there’s a simpler logic:

For the keys used by trainer and listener:

key = "a" → "singleton" or "*args"
key = ("a", "b") → "*args"
key = {"a": "x", "b": "y"} → "**kwargs"
key = (("a", "b"), {"k": "x"}) → "*args, **kwargs"

This is a very straightforward mapping, simpler, more intuitive, and easier to maintain than using signature.

For a select and a key, we can uniquely and directly retrieve the corresponding formatted data.

WDTY?

[blythed]

I thought about it and I’m not entirely sure if the concept of signature is really necessary for the model.

For an individual model, it actually doesn’t seem to need a signature.

The signature only comes into play when we use a trainer or a listener.

In fact, there’s a simpler logic:

For the keys used by trainer and listener:

key = "a" → "singleton" or "*args"
key = ("a", "b") → "*args"
key = {"a": "x", "b": "y"} → "**kwargs"
key = (("a", "b"), {"k": "x"}) → "*args, **kwargs"

This is a very straightforward mapping, simpler, more intuitive, and easier to maintain than using signature.

For a select and a key, we can uniquely and directly retrieve the corresponding formatted data.

WDTY?

Yes makes sense.

[blythed]

I thought about it and I’m not entirely sure if the concept of signature is really necessary for the model.
For an individual model, it actually doesn’t seem to need a signature.
The signature only comes into play when we use a trainer or a listener.
In fact, there’s a simpler logic:
For the keys used by trainer and listener:

key = "a" → "singleton" or "*args"
key = ("a", "b") → "*args"
key = {"a": "x", "b": "y"} → "**kwargs"
key = (("a", "b"), {"k": "x"}) → "*args, **kwargs"

This is a very straightforward mapping, simpler, more intuitive, and easier to maintain than using signature.
For a select and a key, we can uniquely and directly retrieve the corresponding formatted data.
WDTY?

Yes makes sense.

After thinking about this again, the place where this is needed is in the predict_batches function, which is indeed in Model. We could potentially add this as an additional argument to predict_batches. We should also create a new function, Model.predict_many which has the same signature as predict_batches, which can be used as alternative to programming the predict_batches method.