seeding

Lightweight python library for achieving deterministic random seeds

Installation

Currently requires Python 3

pip install seeding

Usage

Basics

Seeding's core usage pattern is to call seeding.seed, supplying both a random seed and any modules for which you want to set the random seed.

import random
import seeding

seeding.seed(0, random)
random.random()
# 0.8444218515250481
seeding.seed(0, random)
random.random()
# 0.8444218515250481

Seeding is compatible with import X as Y syntax and can seed multiple modules at the same time.

import numpy as np
import random
import torch

import seeding

seeding.seed(0, random, np, torch)

Other arguments

The seeding.seed function also takes in an optional keyword argument deterministic. By default, this argument is set to True, but it can be overridden if determinism is not required, which can potentially result in improved performance in some libraries (e.g. PyTorch).

import seeding
import torch

seeding.seed(0, torch, deterministic=False)

Seeding additional objects

For some libraries, like OpenAI Gym, seeding happens at an object level instead of a global level. To seed gym environments, pass both the gym module and any environments you want to seed.

import gym
import seeding

env_name = "CartPole-v0"
env = gym.make(env_name)
other_env = gym.make(env_name)

seeding.seed(0, gym, env, other_env)

Note: the default behavior also seeds the env.action_space; to disable this behavior, use deterministic=False as described above.

Compatibility

Seeding currently works with the following libraries:

• random (built-in module)
• numpy
• gym
• torch
• tensorflow (untested)

If there's another library you'd like it to support, feel free to create an issue for it!