如果不加positioning encoding, transformer效果会差很多。常见的encoding方式有以下几种:
1. 再输入时加固定的encoding
典型的就是Attention Is All You Need中的方法。具体解释不多说,直接放代码:
import math
import torch
def positionalencoding1d(d_model, length):
"""
:param d_model: dimension of the model
:param length: length of positions
:return: length*d_model position matrix
"""
if d_model % 2 != 0:
raise ValueError("Cannot use sin/cos positional encoding with odd dim (got dim={:d})".format(d_model))
pe = torch.zeros(length, d_model)
position = torch.arange(0, length).unsqueeze(1)
div_term = torch.exp((torch.arange(0, d_model, 2, dtype=torch.float) * -(math.log(10000.0) / d_model)))
pe[:, 0::2] = torch.sin(position.float() * div_term)
pe[:, 1::2] = torch.cos(position.float() * div_term)
return pe
def positionalencoding2d(d_model, height, width):
"""
:param d_model: dimension of the model
:param height: height of the positions
:param width: width of the positions
:return: d_model*height*width position matrix
"""
if d_model % 4 != 0:
raise ValueError("Cannot use sin/cos positional encoding with "
"odd dimension (got dim={:d})".format(d_model))
pe = torch.zeros(d_model, height, width)
# Each dimension use half of d_model
d_model = int(d_model / 2)
div_term = torch.exp(torch.arange(0., d_model, 2) *
-(math.log(10000.0) / d_model))
pos_w = torch.arange(0., width).unsqueeze(1)
pos_h = torch.arange(0., height).unsqueeze(1)
pe[0:d_model:2, :, :] = torch.sin(pos_w * div_term).transpose(0, 1).unsqueeze(1).repeat(1, height, 1)
pe[1:d_model:2, :, :] = torch.cos(pos_w * div_term).transpose(0, 1).unsqueeze(1).repeat(1, height, 1)
pe[d_model::2, :, :] = torch.sin(pos_h * div_term).transpose(0, 1).unsqueeze(2).repeat(1, 1, width)
pe[d_model + 1::2, :, :] = torch.cos(pos_h * div_term).transpose(0, 1).unsqueeze(2).repeat(1, 1, width)
return pe
对于1维的encoding可视化如下:
import matplotlib.pyplot as plt
import numpy as np
pe = positionalencoding1d(252*2, 252)
plt.imshow(np.concatenate((pe[:, 0::2], pe[:, 1::2]), axis=1))
plt.colorbar(fraction=0.023, pad=0.05)
plt.show()