|
--- |
|
library_name: keras |
|
tags: |
|
- probabilistic-models |
|
- regression |
|
--- |
|
|
|
## Model description |
|
|
|
This repo contains model weights for the the probabilistic model from [Probabilistic Bayesian Neural Networks](https://keras.io/examples/keras_recipes/bayesian_neural_networks/). This example demonstrates how to build basic probabilistic Bayesian neural networks to account for these two types of uncertainty. We use TensorFlow Probability library, which is compatible with Keras API. |
|
|
|
Taking a probabilistic approach to deep learning allows to account for uncertainty, so that models can assign less levels of confidence to incorrect predictions. Sources of uncertainty can be found in the data, due to measurement error or noise in the labels, or the model, due to insufficient data availability for the model to learn effectively. |
|
|
|
**Full credits go to [Khalid Salama](https://www.linkedin.com/in/khalid-salama-24403144/)** |
|
|
|
## Using this model |
|
This repo contains model weights only. To use this model, refer to the following code contained in load_bnn_model.py. |
|
|
|
## Training and evaluation data 🍷 |
|
|
|
We use the wine quality dataset found [here](https://www.tensorflow.org/datasets/catalog/wine_quality). Each wine was scored from 0-10 by wine experts, and includes 11 physicochemical features about the wine. |
|
|
|
## Versioning |
|
|
|
The training was done using TensorFlow 2.8.0 and TensorFlow Probability 0.16.0. When working with TensorFlow Probability, it is encouraged to check out the [releases](https://github.com/tensorflow/probability/releases/tag/v0.17.0) to make sure you are using a stable TensorFlow counterpart. |
|
|
|
### Training hyperparameters |
|
|
|
| Optimizer | learning_rate | decay | rho | momentum | epsilon | centered | training_precision | |
|
|----|-------------|-----|------|------|-------|-------|------------------| |
|
|RMSprop|0.001|0.0|0.9|0.0|1e-07|False|float32| |
|
|
|
|
|
|
