bartz.mcmcstep.AutoBatchedReduction¶
- class bartz.mcmcstep.AutoBatchedReduction(min_batch_size=128.0, beta_sm=48.0, gamma=0.4, batches_inner=True, contiguous=False)[source]¶
BatchedReductionthat picksnum_batchesautomatically per platform.A flat target on cpu, and on gpu a count scaling with the SM count and the multivariate outcome size. Only cpu and cuda are supported; any other platform raises at lowering.
- min_batch_size: float = 128.0¶
caps the batch count at
n / min_batch_size.- Type:
Minimum datapoints per batch on gpu
- beta_sm: float = 48.0¶
the batch count saturates at
beta_sm * n_sms * m ** -gamma, withn_smsthe gpu’s SM count andmthe multivariate work per datapoint.- Type:
Batches per streaming multiprocessor on gpu
- gamma: float = 0.4¶
Exponent by which multivariate outcomes (
mvalues per datapoint) shrink the saturation batch count on gpu.
- batches_inner: bool = True¶
Same as
BatchedReduction.
- contiguous: bool = False¶
Same as
BatchedReduction.