neuralnetlib 3.3.7

A flexible deep learning framework built from scratch using only NumPy

Neuralnetlib

📝 Description

This is a handmade deep learning framework library, made in python, using numpy as its only external dependency.

I made it to challenge myself and to learn more about deep neural networks, how they work in depth.

The big part of this project, meaning the Multilayer Perceptron (MLP) part, was made in a week.

I then decided to push it even further by adding Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Autoencoders, Variational Autoencoders (VAE) and Transformers (WIP).

This project will be maintained as long as I have ideas to improve it, and as long as I have time to work on it.

📦 Features

• Many models architectures (sequential, functional, autoencoder) 🏗
• Many layers (dense, dropout, conv1d/2d, pooling1d/2d, flatten, embedding, batchnormalization, textvectorization, lstm, gru, attention and more) 🧠
• Many activation functions (sigmoid, tanh, relu, leaky relu, softmax, linear, elu, selu) 📈
• Many loss functions (mean squared error, mean absolute error, categorical crossentropy, binary crossentropy, huber loss) 📉
• Many optimizers (sgd, momentum, rmsprop, adam) 📊
• Supports binary classification, multiclass classification, regression and text generation 📚
• Preprocessing tools (tokenizer, pca, ngram, standardscaler, pad_sequences, one_hot_encode and more) 🛠
• Callbacks and regularizers (early stopping, l1/l2 regularization) 📉
• Save and load models 📁
• Simple to use 📚

⚙️ Installation

You can install the library using pip:

pip install neuralnetlib

💡 How to use

Basic usage

See this file for a simple example of how to use the library.
For a more advanced example, see this file for using CNN.
You can also check this file for text classification using RNN.

Advanced usage

See this file for an example of how to use VAE to generate new images.
And this file for an example of how to generate new dinosaur names.

More examples in this folder.

You are free to tweak the hyperparameters and the network architecture to see how it affects the results.

🚀 Quick examples (more here)

Binary Classification

from neuralnetlib.models import Sequential
from neuralnetlib.layers import Input, Dense
from neuralnetlib.activations import Sigmoid
from neuralnetlib.losses import BinaryCrossentropy
from neuralnetlib.optimizers import SGD
from neuralnetlib.metrics import accuracy_score

# ... Preprocess x_train, y_train, x_test, y_test if necessary (you can use neuralnetlib.preprocess and neuralnetlib.utils)

# Create a model
model = Sequential()
model.add(Input(10))  # 10 features
model.add(Dense(8))
model.add(Dense(1))
model.add(Activation(Sigmoid()))  # many ways to tell the model which Activation Function you'd like, see the next example

# Compile the model
model.compile(loss_function='bce', optimizer='sgd')

# Train the model
model.fit(X_train, y_train, epochs=10, batch_size=32, metrics=['accuracy'])

Multiclass Classification

from neuralnetlib.activations import Softmax
from neuralnetlib.losses import CategoricalCrossentropy
from neuralnetlib.optimizers import Adam
from neuralnetlib.metrics import accuracy_score

# ... Preprocess x_train, y_train, x_test, y_test if necessary (you can use neuralnetlib.preprocess and neuralnetlib.utils)

# Create and compile a model
model = Sequential()
model.add(Input(28, 28, 1)) # For example, MNIST images
model.add(Conv2D(32, kernel_size=3, padding='same'), activation='relu')  # activation supports both str...
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=2))
model.add(Dense(64, activation='relu'))
model.add(Dense(10, activation=Softmax()))  # ... and ActivationFunction objects
model.compile(loss_function='categorical_crossentropy', optimizer=Adam())


model.compile(loss_function='categorical_crossentropy', optimizer=Adam())  # same for loss_function and optimizer

# Train the model
model.fit(X_train, y_train_ohe, epochs=5, metrics=['accuracy'])

Regression

from neuralnetlib.losses import MeanSquaredError
from neuralnetlib.metrics import accuracy_score

# ... Preprocess x_train, y_train, x_test, y_test if necessary (you can use neuralnetlib.preprocess and neuralnetlib.utils)

# Create and compile a model
model = Sequential()
model.add(Input(13))
model.add(Dense(64, activation='leakyrelu'))
model.add(Dense(1), activation="linear")

model.compile(loss_function="mse", optimizer='adam')  # you can either put acronyms or full name

# Train the model
model.fit(X_train, y_train, epochs=100, batch_size=128, metrics=['accuracy'])

You can also save and load models:

# Save a model
model.save('my_model.json')

# Load a model
model = Model.load('my_model.json')

📜 Output of the example file

Here is the decision boundary on a Binary Classification (breast cancer dataset):

[!NOTE] PCA (Principal Component Analysis) was used to reduce the number of features to 2, so we could plot the decision boundary. Representing n-dimensional data in 2D is not easy, so the decision boundary may not be always accurate. I also tried with t-SNE, but the results were not good.

Here is an example of a model training on the mnist using the library

Here is an example of a loaded model used with Tkinter:

Here, I decided to print the first 10 predictions and their respective labels to see how the network is performing.

Here is the generated dinosaur names using a simple RNN and a list of existing dinosaur names.

You can of course use the library for any dataset you want.

✏️ Edit the library

You can pull the repository and run:

pip install -e .

And test your changes on the examples.

🎯 TODO

• Add more model architecture support (like transformers, gan, etc)
• Add support for stream dataset loading to allow loading large datasets (larger than your RAM)
• Add more callbacks
• Add more layers
• Add more preprocessing tools
• Add BPE (Byte Pair Encoding) tokenizer
• Add cuDNN support to allow the use of GPUs
• Visual updates (tabulation of model.summary() parameters calculation, colorized progress bar, etc.)

🐞 Know issues

• The transformer has gradient issues (normalization and often constant attention weights after a few epochs)
• The save feature of some very rare cases (layers/models) aren't working properly (I just need to read my old code again)

✍️ Authors

• Marc Pinet - Initial work - marcpinet