mirror of
https://github.com/apachecn/ailearning.git
synced 2026-02-10 22:05:11 +08:00
71 lines
1.9 KiB
Markdown
71 lines
1.9 KiB
Markdown
# Theano 随机数流变量
|
||
|
||
In [1]:
|
||
|
||
```py
|
||
import theano
|
||
import theano.tensor as T
|
||
import numpy as np
|
||
|
||
```
|
||
|
||
```py
|
||
Using gpu device 1: Tesla C2075 (CNMeM is disabled)
|
||
|
||
```
|
||
|
||
`Theano` 的随机数变量由 `theano.sandbox.rng_mrg` 中的 `MRG_RandomStreams` 实现（`sandbox` 表示是实验代码）：
|
||
|
||
In [2]:
|
||
|
||
```py
|
||
from theano.sandbox.rng_mrg import MRG_RandomStreams
|
||
|
||
```
|
||
|
||
新建一个 `MRG_RandomStreams(seed=12345, use_cuda=None)` 实例：
|
||
|
||
In [3]:
|
||
|
||
```py
|
||
srng = MRG_RandomStreams()
|
||
|
||
```
|
||
|
||
它支持以下方法：
|
||
|
||
* `normal(size, avg=0.0, std=1.0, ndim=None, dtype=None, nstreams=None)`
|
||
* 产生指定形状的、服从正态分布 $N(avg, std)$ 的随机数变量，默认为标准正态分布
|
||
* `uniform(size, low=0.0, high=1.0, ndim=None, dtype=None, nstreams=None)`
|
||
* 产生指定形状的、服从均匀分布 $U(low, high)$ 的随机数变量，默认为 0-1 之间的均匀分布
|
||
* `binomial(size=None, n=1, p=0.5, ndim=None, dtype='int64', nstreams=None)`
|
||
* 产生指定形状的、服从二项分布 $B(n,p)$ 的随机数变量
|
||
* `multinomial(size=None, n=1, pvals=None, ndim=None, dtype='int64', nstreams=None)`
|
||
* 产生指定形状的、服从多项分布的随机数变量
|
||
|
||
与 np.random.random 不同，它产生的是随机数变量，而不是随机数数组，因此可以将 `size` 作为参数传给它：
|
||
|
||
In [4]:
|
||
|
||
```py
|
||
rand_size = T.vector(dtype="int64")
|
||
|
||
rand_normal = srng.normal(rand_size.shape)
|
||
rand_uniform = srng.uniform(rand_size.shape)
|
||
rand_binomial = srng.binomial(rand_size.shape)
|
||
|
||
f_rand = theano.function(inputs = [rand_size],
|
||
outputs = [rand_normal, rand_uniform, rand_binomial])
|
||
|
||
print f_rand(range(5))[0]
|
||
print f_rand(range(5))[1]
|
||
print f_rand(range(5))[2]
|
||
|
||
```
|
||
|
||
```py
|
||
[ 0.10108768 -1.64354193 0.71042836 -0.77760422 0.06291872]
|
||
[ 0.23193923 0.71880513 0.03122572 0.97318739 0.99260223]
|
||
[0 1 0 1 1]
|
||
|
||
``` |