Project for Bayesian Statistics Class.
In this repository, we implement two approaches for introducing uncertainty in neural networks based on the code available in tensorflow_probability examples.
frequentist.py implements a frequentist approach. It can be launched from the terminal by simply choosing the right environment containing the following packages: seaborn, numpy, tensorflow, pyplot. The command is the following:
python frequentist.py --learning_rate --max_steps --batch_size --data_dir --model_dir --viz_steps --print_steps --run_steps
Default values are those written in the associated report.
bayesian.py implements a bayesian variational approach. It can be launched from the terminal by simply choosing the right environment containing the following packages: seaborn, numpy, tensorflow_probability, pyplot. The command is the following:
python bayesian.py --learning_rate --max_steps --batch_size --data_dir --model_dir --viz_steps --print_steps --num_monte_carlo
Default values are those written in the associated report.
WARNING
The --fake_data flag should always be True which is the default value. Program will not run otherwise.