Wiener filtering flat CMB maps with a Neural Network

This is the code for the paper "Fast Wiener filtering of CMB maps with Neural Networks" (https://arxiv.org/abs/1905.05846), accepted to the NeurIPS 2019 workshop "Machine Learning and the Physical Sciences".

Requirements:

• tensorflow (tested on python 2.7 and tf 1.11)
• quicklens (https://github.com/dhanson/quicklens)

Environment variables:

• Make sure PYTHONPATH includes the mlcmb folder.

To run an example:

1. Edit the config mlcmb/config/config_128_t_35muK.ini. There set the datapath to where you want to store the datasets and networks. Create the two subfolders mentioned in the config.

2. Create a data set for this config by running python trainingdata.py configs/config_128_t_35muK.ini. This takes a while as it will Wiener filter the test data also, with conjugate gradient.

3. Train on the data by running python train.py configs/config_128_t_35muK.ini

4. After training generate evaluation metrics python eval.py configs/config_128_t_35muK.ini

5. View quality control plots with the notebook notebooks/wiener_results_t.ipynb