Quick Start

Requirements : - Docker

Start by building the python image with docker compose :

docker-compose up

Note

We’re assuming that you’re running the commands from the root of the project.

Access the docker service which will act as your working environment :

docker-compose run python-example

To know if you’re in you should see a change in the cli’s prompt :

root@ad245b0ff5c4:/app#

Now all you have to do is navigate to the python example project of your liking and run the scripts just like you would normally. The following example is for running the MNIST example which consists of downloading the handwritten digits dataset. Then, creating a Neural Network model and training it on the dataset.

cd examples/train-predict-MNIST && mkdir dataset && python main.py

If you look at the main.py file, you’ll notice :

  • Fetching the MNIST handwritten digits dataset .. code-block:: python

    if not file_exists(MNIST_DATASET_FILE):

    print(“Mnist dataset not found”) get_MNIST_dataset(MNIST_DATASET_FILE)

  • Initiating and composing a Neural Network

network = NNP.models.Network()

network.addLayer(NNP.layers.Flatten((28, 28)))
network.addLayer(NNP.layers.Dense(128, NNP.ACTIVATION.RELU, NNP.WEIGHT_INIT.HE))
network.addLayer(NNP.layers.Dense(10, NNP.ACTIVATION.SOFTMAX, NNP.WEIGHT_INIT.LECUN))

# Setting up the networks parameters
network.setup(optimizer=NNP.optimizers.Adam(0.01), loss=NNP.LOSS.MCE)

  • Formatting and normalizing the data and then storing it in a TrainingData3dI object. Which simplifies batching.

# preparing the training data
f_x_train = [normalize_img(x) for x in x_train]

trainingData = NNP.TrainingData3dI(f_x_train[:NUM_TRAININGS], y_train[:NUM_TRAININGS])

trainingData.batch(128) # Creating batches of 128 inputs

  • Training the data

  • Computing the accuracy

  • Serializing the trained model and saving it in a binary format

# Saving the trained model in a bin file
NNP.models.Model.save_to_file('./model.bin', network)

saved_model = NNP.models.Network()

  • Loading the model into a new instance of Network from the file again (simply for showcase)

# Saving the trained model in a bin file
NNP.models.Model.save_to_file('./model.bin', network)

saved_model = NNP.models.Network()

NNP.models.Model.load_from_file('./model.bin', saved_model)

  • Testing the model with the test_data

Installation

Warning

The project must be built with cmake in order to use the python library.

Then inside your python script, you must include the module path to the system path :

import sys
sys.path.append('/path/to/build/folder')

Then you can import the module just like any other :

import NeuralNetPy as NNP

Attention

The path to the build folder must be set before importing the module.