This is code for the Neural Moving Average Model paper.

• TensorFlow 1.8.0
• NumPy 1.14.5

The folder dat contains data generated by AR_dat_gen. This data is then used by AR.py which trains a model. Both of these python files can be called by main.py. main.py uses the hyperparameters defined in hyperparameters.txt to (a) generate synthetic data and then (b) fit the model to that synthetic data.

After downloading the source, inference for the AR(1) process can be performed using the default values (as listed in the paper):

python main.py hyperparameters.txt

This uses TensorBoard to visualise training losses and parameter posteriors. Data for this purpose is saved into a directory called 'train'. To visualise this in TensorBoard call (in your working directory):

tensorboard --logdir=train/

and navigate to localhost:6006 in your browser.

The data-generation process and/or the inference proceudre can be modified by either editing the values in the hyperparameter text file or appending command line options. The command line options are detailed as follows:

The code for the other experiments can be run using the lotka_volterra_partial.py, fitz_nag_NVP.py and SV_dense.py scripts.