UCCRecalibration fit method running on one CPU core

[AdelNamani]

UCCRecalibration.fit runs on only one CPU core.
Is there a way to parallelize the execution of this method?

[pronics2004]

@jirin1a do you have thoughts on this?

[jirin1a]

@AdelNamani, yes, the fit() method currently uses a single processor. Algorithmically, it can be parallelized. Before we look into this, can you give us more detail: what is the size of the data object with which you are invoking fit() and the amount of data you are recalibrating?

[AdelNamani]

Here are some details about the size of the data object I am trying to train on:

Int64Index: 229297 entries, 
Columns: 605 entries, 
dtypes: float64(84), int64(3), object(3), uint64(1), uint8(514)
memory usage: 273.3+ MB

This represents 25% of my train data.
Executing the UCCRecalibration.fit takes approximately 2 hours.

[jirin1a]

I see. We may put this request on our todo list - speeding up the fit() call make sense.
In the meantime, to obtain better runtimes, subsampling the data used in the fit call may be a suitable alternative. From the instance naming, I am surmising that after the fit() call you are rescaling all your data to minimize a cost (?). If so, the data object, say X, can be randomly subsampled to, say, 10% of the original size with the resulting optimum applied to the full batch of data.

[AdelNamani]

That's what I am doing actually, thanks!