amix / neuralist

A Python interface to access neural-redis

neuralist

A Python interface to access neural-redis, a neural networks module for Redis.

🚨 This stuff is an alpha stage (like neural-redis), so use it with caution!

Installation 🚀

To install simply do following:

$ pip install neuralist

Example usage

from neuralist import NeuralNetwork

nn = NeuralNetwork('additions',
                   type='regressor',
                   inputs=('number1', 'number2'),
                   outputs=('result',),
                   hidden_layers=(3, ),
                   dataset_size=50,
                   testset_size=10)

nn.observe_train(input={'number1': 1, 'number2': 1},
                 output={'result': 2})
                
nn.train()

while nn.is_training():
    print 'Training...'
    time.sleep(1)

print nn.run({'number': 1, 'number': 2})

API spec

NeuralNetwork class

The NeuralNetwork object follows the arguments of the nn.create command.

Required arguments:

• name: (string) The name of the neural network
• type: (string) Type of network, can be 'classifier' or 'regressor'
• inputs: (tuple) The keys of the input dict
• outputs: (tuple) The keys of the output dict
• hidden_layers: (tuple) zero or more arguments indicating the number of hidden units, one number for each layer.
• dataset_size: (int) Max number of data samples in the training dataset
• testset_size: (int) Max number of data samples in the testing dataset

Optional arguments.

• normalize: (bool) Default True. Specify if you want the network to normalize your inputs. Use this if you don't know
• redis_client: (Redis.Client) Specify which Redis client to use
• prefix: (string) Default nn:. Ability to prefix the neural networks, so they don't collide with other keys
• auto_create: (bool) Default True. Neural network will be auto created if it isn't found

NeuralNetwork.run(input)

Run the network on the input, returning a dict as the result.

Required arguments:

• input: (dict) The input vector encoded as dict

Example:

print(nn.run({'number1': 1, 'number2': 2}))
# =>
{'result': '3.7173097692441601'}

NeuralNetwork.classify(input)

Run the network on the input, returning a class as the result.

Required arguments:

• input: (dict) The input vector encoded as dict

Example (from Titanic example):

print(nn.classify({...}))
# =>
'dead'

NeuralNetwork.observe_train(input, output)

Add a data sample into the training dataset.

Required arguments:

• input: (dict) The input vector encoded as dict
• output: (dict) The output vector encoded as output

Example:

nn.observe_train(input={'number1': 1, 'number2': 2},
                 output={'result': 3})

nn.observe_train(input={'age': 25, 'male': 0, 'female': 1},
                 output={'dead': 0, 'alive': 1})

NeuralNetwork.observe_test(input, output)

Add a data sample into the testing dataset.

Required arguments:

• input: (dict) The input vector encoded as dict
• output: (dict) The output vector encoded as output

NeuralNetwork.is_training()

Returns True if the network is being trained.

NeuralNetwork.train()

Train a network in a background thread.

Optional arguments.

• max_cycles: (int) Default None. Max cycles count
• max_time: (int) Default None. Max time to train
• autostop: (bool) Default True. If no autostop is True, trains the network till the maximum number of cycles or milliseconds are reached. If no maximum number of cycles is specified there are no cycles limits. If no milliseconds are specified, the limit is set to 10 seconds.
• backtrack: (bool) Default True. If backtrack is True, and autostop is also True, while the network is trained, the trainer thread saves a copy of the neural network every time it has a better score compared to the previously saved one and there are hints suggesting that overfitting may happen soon. This network is used later if it is found to have a smaller error.

NeuralNetwork.is_created()

Returns True if the neural network is created.

NeuralNetwork.create()

Create the neural network.

NeuralNetwork.recreate()

Recreate the neural network.

NeuralNetwork.delete()

Delete the neural network.

NeuralNetwork.info()

Returns a dict with info about the neural network. The keys are the same that are found on nr.info command.

More examples

To see how to use this library please see the tests directory -- there's an example of a neural network that can do additions and one that can answer if a Titanic passenger will survive or not.

To run the examples:

$ python -m tests.test_addition
$ python -m tests.test_titanic