11

As I understand the syntax is 

In[88]: np.random.seed(seed=0)

In[89]: np.random.rand(5) < 0.8
Out[89]: array([ True,  True,  True,  True,  True], dtype=bool)
In[90]: np.random.rand(5) < 0.8
Out[90]: array([ True,  True, False, False,  True], dtype=bool)

However, when I run the rand(), I get different results. Is there something I am missing with the seed function?

Zizouz212
  • 4,908
  • 5
  • 42
  • 66
unj2
  • 52,135
  • 87
  • 247
  • 375
  • 5
    Yes; if you want the same result every time you need to call `seed` in-between. It's already deterministic, but that *doesn't* mean you get the same result every time you call `np.random`. – jonrsharpe Sep 27 '15 at 15:39
  • 1
    This may also be useful for your purposes: [Difference between `np.random.seed()` and `np.random.RandomState()`](http://stackoverflow.com/a/22994890/1730674) – askewchan Sep 27 '15 at 18:15

1 Answers1

12

Think of a generator:

def gen(start):
    while True:
        start += 1
        yield start

That will continuously give the next number from the number you insert to the generator. With seeds, it's nearly the same concept. I try to set a variable in which to generate data from, and the position in within that is still saved. Let's put this into practice:

>>> generator = gen(5)
>>> generator.next()
6
>>> generator.next()
7

If you want to restart, you need to restart the generator as well:

>>> generator = gen(5)
>>> generator.next()
6

The same idea with the numpy object. If you want the same results over time, you need to restart the generator, with the same arguments.

>>> np.random.seed(seed=0)
>>> np.random.rand(5) < 0.8
array([ True,  True,  True,  True,  True], dtype=bool)
>>> np.random.rand(5) < 0.8
array([ True,  True, False, False,  True], dtype=bool)
>>> np.random.seed(seed=0) # reset the generator!
>>> np.random.rand(5) < 0.8
array([ True,  True,  True,  True,  True], dtype=bool)
Zizouz212
  • 4,908
  • 5
  • 42
  • 66