如何在PyTorch中将numpy数组转换为LongTensor类型的变量？

Question

I am trying to convert numpy array into PyTorch LongTensor type Variable as follows:

import numpy as np
import torch as th

y = np.array([1., 1., 1.1478225, 1.1478225, 0.8521775, 0.8521775, 0.4434675])
yth = Variable(th.from_numpy(y)).type(torch.LongTensor)

However the result I am getting is a rounded off version:

tensor([ 1,  1,  1,  1,  0,  0,  0])

How can I keep the precision of numpy array while getting LongTensor variable?

Expected result should be:

tensor([1., 1., 1.1478225, 1.1478225, 0.8521775, 0.8521775, 0.4434675])

Answer 1

LongTensor represents tensors with values of type long / int64 (cf table in doc ). Your float values are thus converted (ie rounded) to integers.

To keep float values, use FloatTensor ( float32 ) or DoubleTensor ( float64 ) instead.