how to

x shape: (n, ns)

Aug 15, 2024
notesjulyfun技术学习d2l
2 Minutes
208 Words

拉普拉斯平滑

在统计单词 / 连续单词出现次数后,计算出现概率 $P$ 时,添加一个超参数小常量。效果:若常量趋于无穷大,则概率为 $1 / “单词总数”$(对于连续单词 AB,则是趋于 $1 / P(“A”)$)

齐普夫定律

分布满足对数坐标系上的下降直线。一元语法,n 元语法均遵守这个分布。

构造的数据集形如

1
for X, Y in seq_data_iter_sequential(my_seq, batch_size=2, num_steps=5):
2
print('X: ', X, '\nY:', Y)
1
# x shape: (n, ns)
2
# y shape: (n, ns)
3
X: tensor([[ 2, 3, 4, 5, 6],
4
[18, 19, 20, 21, 22]])
5
Y: tensor([[ 3, 4, 5, 6, 7],
6
[19, 20, 21, 22, 23]])
7
X: tensor([[ 7, 8, 9, 10, 11],
8
[23, 24, 25, 26, 27]])
9
Y: tensor([[ 8, 9, 10, 11, 12],
10
[24, 25, 26, 27, 28]])
11
X: tensor([[12, 13, 14, 15, 16],
12
[28, 29, 30, 31, 32]])
13
Y: tensor([[13, 14, 15, 16, 17],
14
[29, 30, 31, 32, 33]])

44:25

Article title:x shape: (n, ns)
Article author:Julyfun
Release time:Aug 15, 2024
Copyright 2025
Sitemap