DeepThink

DeepThink is a deep learning library for Python, designed as a learning project and as a resource for others looking to learn about deep learning. It provides a high-level interface for building, training, and evaluating deep learning models, as well as a range of utilities for working with data and optimizing models.

Features

• A high-level API for defining, training, and evaluating models with minimal code
• Utilities for data loading, preprocessing, and model evaluation
• Tools for debugging, profiling, and optimizing models
• A range of examples to illustrate several use cases

Please note that DeepThink is a basic deep learning library and may not have the same level of performance or support for advanced features as other more established deep learning libraries.

Installation

pip install deepthink

Quickstart

Here is a simple example of how to use DeepThink to train a deep learning model:

from deepthink.optimizers import Adam
from deepthink.layers import Dense, Conv2D, MaxPooling, Flatten
from deepthink.model import Model
from deepthink.activations import ReLU,  Softmax
from deepthink.utils import load_mnist_data
from deepthink.loss import CategoricalCrossEntropy

# Load dataset
training_data, test_data = load_mnist_data()

# Creating a model
optimizer = Adam(0.001)
model = Model(optimizer, cost=CategoricalCrossEntropy(), batch_size=64)

model.add_layer(Conv2D(kernel_size=5, n_filters=8,
                       input_shape=(64, 1, 28, 28)))
model.add_layer(ReLU())
model.add_layer(MaxPooling())
model.add_layer(Flatten())
model.add_layer(Dense(16))
model.add_layer(ReLU())
model.add_layer(Dense(10))
model.add_layer(Softmax())

model.initialize()

# Train the model
history = model.train(training_data, test_data, epochs=5)

You can find additional examples in the examples directory.

Contributing

Contributions are more than welcome to DeepThink! If you would like to report a bug, request a feature, or contribute code, please create an issue or submit a pull request.

License

DeepThink is released under the MIT License.