Chronos : List the input data type of forecaster and locate some dismatch

[rnwang04]

Nowdays, input data types for different methods in BaseForecaster and AutoformerForecaster is different and is confusing. The following table lists the support data types of each method:

Method	numpy input	loader input	tsdataset instance	xshard input
BaseForecaster.fit	✔	✔	✔	✔
AutoformerForecaster.fit	✔	✔	❌	❌
BaseForecaster.tune	✔	❌	❌	❌
AutoformerForecaster.tune	✔	❌	❌	❌
BaseForecaster.predict	✔	✔	✔	✔
AutoformerForecaster.predict(*)	✔	✔	❌	❌
BaseForecaster.predict_with_onnx	✔	✔	✔	❌
AutoformerForecaster.predict_with_onnx(×)	-	-	-	-
BaseForecaster.predict_with_openvino	✔	❌	❌	❌
AutoformerForecaster.predict_with_onnx(×)	-	-	-	-
BaseForecaster.evaluate	✔	✔	✔	✔
AutoformerForecaster.evaluate(*)	✔	✔	❌	❌
BaseForecaster.predict_interval	✔	✔	✔	❌
AutoformerForecaster.predict_interval	✔	✔	❌	❌

Note: × means don't have this method, * means there maybe some dismatch in metrics. For fit and predict_interval, there are two parameters : data and validation_data(or val_data)

Method	numpy input	loader input	tsdataset instance	xshard input
data for BaseForecaster.fit	✔	✔	✔	✔
validation_data for BaseForecaster.fit	✔	✔	❌	❌
data for AutoformerForecaster.fit	✔	✔	❌	❌
validation_data for AutoformerForecaster.fit(x)	-	-	-	-
data for BaseForecaster.predict_interval	✔	✔	✔	❌
val_data for BaseForecaster.predict_interval	✔	✔	✔	❌
data for AutoformerForecaster.predict_interval	✔	✔	❌	❌
val_data for AutoformerForecaster.predict_interval	✔	✔	❌	❌

Note: AutoformerForecaster don't support validation_data in fit now (#5865)

some methods in BaseForecaster but don't in Autoformer : get_model() (#5865)

In next (several) PR, I may make it becomes:

Method	numpy input	loader input	tsdataset instance	xshard input
BaseForecaster.fit	✔	✔	✔	✔
AutoformerForecaster.fit	✔	✔	✔(#5846)	❌
BaseForecaster.tune	✔	❌	❌	❌
AutoformerForecaster.tune	✔	❌	❌	❌
BaseForecaster.predict	✔	✔	✔	✔
AutoformerForecaster.predict(*)	✔	✔	✔(#5846)	❌
BaseForecaster.predict_with_onnx	✔	✔	✔	❌
AutoformerForecaster.predict_with_onnx(×)	-	-	-	-
BaseForecaster.predict_with_openvino	✔	✔(#5846)	✔(#5846)	❌
AutoformerForecaster.predict_with_onnx(×)	-	-	-	-
BaseForecaster.evaluate	✔	✔	✔	✔
AutoformerForecaster.evaluate(*)	✔	✔	✔(#5846)	❌(?)
BaseForecaster.predict_interval	✔	✔	✔	❌
AutoformerForecaster.predict_interval	✔	✔	✔(#5846)	❌

Method	numpy input	loader input	tsdataset instance	xshard input
data for BaseForecaster.fit	✔	✔	✔	✔
validation_data for BaseForecaster.fit	✔	✔	✔(#5516)	✔(#5846)
data for AutoformerForecaster.fit	✔	✔	✔(#5846)	❌
validation_data for AutoformerForecaster.fit(x)	✔(#5865)	✔(#5865)	✔(#5865)	❌
data for BaseForecaster.predict_interval	✔	✔	✔	❌
val_data for BaseForecaster.predict_interval	✔	✔	✔	❌
data for AutoformerForecaster.predict_interval	✔	✔	✔(#5846)	❌
val_data for AutoformerForecaster.predict_interval	✔	✔	✔(#5846)	❌

[rnwang04]

@TheaperDeng How do you think of this?

[TheaperDeng]

@TheaperDeng How do you think of this?

yeah, I think that should be reasonable to first make sure numpy, loader and tsdatset is supported. I think you may split it into 2 PRs since there are quite a few support we need to do here.

[TheaperDeng]

Some of them is on-going in this PR, I will handle them: https://github.com/intel-analytics/BigDL/pull/5516