Closed liamgd closed 1 year ago
out = pandas.DataFrame(
torch.cat(results),
index=index_chunk,
columns=columns,
).astype(float)
out = pandas.DataFrame(
torch.cat(results),
index=index_chunk,
columns=columns,
)
out = pandas.DataFrame(
torch.cat(results).cpu().numpy(),
index=index_chunk,
columns=columns,
)
The fastest method:
out = pandas.DataFrame(
torch.cat(results).cpu().numpy(),
index=index_chunk,
columns=columns,
)
Was implemented in 97c456d.
The regression_full algorithm utilizes the power of the CUDA device much better than regression_single and has a large number of optimizations that drastically reduce the time it takes to compute the regression result tensors. However, the algorithm is severely bottlenecked by the saving of these tensors, particularly on the GPU because of the tensor conversion from GPU to CPU.