TensorBase provides a set of Python classes that abstract the typical functions involved in a machine learning application and facilitate the creation of new applications from data management and to model implementation. TensorBase differs from other TensorFlow-compatible APIs such as Keras and PrettyTensor in several ways:
- Has a simpler structure but demands more knowledge of TensorFlow syntax
- Facilitates the creation of new layers and functions in networks and training
- Addresses data and metrics (e.g. loss, accuracy) management
The TensorBase packages contains 3 Classes in base.py and 1 Class in data.py.
- Model: a parent class that defines the general structure of TensorFlow models and manages metrics.
- Layers: a parent class that iteratively creates connected and convolutional networks.
- Data: a parent class for batch generation.
- MNIST: a child class that generates batchs for the MNIST dataset.