tinynet
A tiny neural network library
No training
This library provides no training algorithm. You can easily set up a neuroevolution framework by including any black-box search algorithm, from RWG (see example below) to CMA-ES.
Installation
pip install tinynet
Usage
from tinynet import RNN1L
import numpy as np
net_struct = [3, 5, 2]
net = RNN(net_struct) # try also FFNN
net.set_weights(np.random.randn(net.nweights))
out = net.activate(np.zeros(net.ninputs))
assert len(out) == net.noutputs
assert len(net.state) == net.ninputs + 1 + net.noutputs # input, bias, recursion
Neuroevolution application on the OpenAI Gym
Check out this GitHub gist to run the Bipedal Walker using pre-trained weights.
The example below tackles the CartPole from scratch using RWG.
import numpy as np
import tinynet
import gym # just `pip install gym`
from time import sleep # slow down rendering
# Environment setup
env = gym.make("CartPole-v1")
# env = gym.wrappers.Monitor(env, 'video', force = True) # Uncomment to save video
# Get input size and output size from the environment
nactions = env.action_space.n
ninputs = env.reset().size
# Hidden layers are arbitrarily added
# hidden = [20, 10, 20]
hidden = [] # ... but unnecessary with the CartPole
net_struct = [ninputs, *hidden, nactions]
# Network setup is straightforward (defaults: `act_fn=np.tanh, init_weights=None`)
net = tinynet.FFNN(net_struct) # also try `RNN(net_struct)`
# Get random seed for deterministic fitness (for simplicity)
rseed = np.random.randint(1e10)
# Fitness function: gameplay loop
def fitness(ind, render=False):
env.seed(rseed) # makes fitness deterministic
obs = env.reset()
score = 0
done = False
net.set_weights(ind)
while not done:
if render:
env.render()
sleep(0.5)
action = net.activate(obs).argmax()
obs, rew, done, info = env.step(action)
score += rew
if render: env.render() # render last frame
print(f"Score: {score}")
return score
# RWG does not distinguish between populations and generations
max_ninds = 1000
# Neuroevolution (RWG) loop
for nind in range(max_ninds):
ind = np.random.randn(net.nweights)
score = fitness(ind)
if score >= 195:
print(f"Game solved in {nind} trials")
break
# Replay winning individual
fitness(ind, render=True)
# You may want to drop into a console here to examine the results
import IPython; IPython.embed()
Metadata
Release files for tinynet 0.1.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| tinynet-0.1.2.tar.gz | 5.3 kB | Without provenance | Details for tinynet-0.1.2.tar.gz |
Release files / tinynet-0.1.2.tar.gz
| Download URL | tinynet-0.1.2.tar.gz |
|---|---|
| Size | 5.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
6dfe4a75891a6e05a2932577bafbb92981934219198e897bc18ebc00926c3bca
|
|
BLAKE2b-256 checksum How to use checksums |
2b13175970b0cbacd36722d5c70ad585b857c684cc5a874b9751e9d8551770f9
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.1.1 pkginfo/1.5.0.1 requests/2.23.0 setuptools/46.4.0 requests-toolbelt/0.9.1 tqdm/4.46.0 CPython/3.6.9
|
Security for tinynet 0.1.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.