Can we change the bin size in pipeline for tokenizing during inference?

[sriyachakravarthy]

Hi!
I would like to know if we can work on high frequency data, if so, what is the maximum frequency?
Can we reduce the bi size for tokenizing during inference ?
Thanks!

[lostella]

@sriyachakravarthy the models were trained on data down to 5-minute granularity.

No, it's not possible to change the bin size during inference: the bin size used for quantization is deeply tied to how the model represents real-valued data internally, so changing it would break the model.

[sriyachakravarthy]

@lostella Thank you!

I see!
So, my understanding is that the minimum time difference between the samples is expected to be 5 min?
And if we have data samples much faster than that, should we re-train the models?

[sriyachakravarthy]

or interpolate the data?