@gganapavarapu Looks like it is failing because we are passing ID columns and this is not yet supported for Chronos?

Something is up with the freq parameter too.

@gganapavarapu Looks like it is failing because we are passing ID columns and this is not yet supported for Chronos?

Correct, ID columns is not supported. But, I was using only target columns in chronos handler. So, this should not be problem. [10] not found in axis is because the input data length for chronos is 1 as minimum_context_length is set to 1, and we are trying to drop this row in the input data in our tests.

TimeSeriesProcessor is not able to derive frequency as input data length is 1 in chronos lib tests. Chronos service tests are passing. So, either we need to handle this in TSFM service components or increase minimum_context_length in chronos tsfm_config.json. Our tests are expecting minimum of length 11.

Running few tests to evaluate the options.

Regarding increasing the minimum context length, that seems reasonable to me but @wgifford are there common use cases out there where someone would expect to get a forecast with a very small context like 1? I can see this for more traditional approaches like Arima but for DL-based models it seems like a degenerate case to me.

converting this PR to draft until all units are passed.

Merged main branch into chrono_bolt branch. This requires review of all the files from merge.

Regarding increasing the minimum context length, that seems reasonable to me but @wgifford are there common use cases out there where someone would expect to get a forecast with a very small context like 1? I can see this for more traditional approaches like Arima but for DL-based models it seems like a degenerate case to me.

minimum context length is increased to 16. Tests are passing now without changes. Thank you @wgifford .

This PR supports ID columns and model specific parameters for chronos models.