Fix Example 2 installation problems

02-sgd/.gitignore

@@ -0,0 +1,3 @@
+data
+runs
+.snakemake

02-sgd/README.md

@@ -2,16 +2,64 @@
 
 This is a complete example demonstrating how MPoL works using simulated data.
 
-Before starting, you should have already run the scripts in the `generate-mock-baselines` folder to produce a mock sky image and interferometer baselines in a file called `mock_data.npz`. Then, you should copy that file to this repository under `data/mock_data.npz`.
+# Prerequisites
 
-This repository assumes that you will run all scripts from this `sgd` directory (the one containing `sgd/README.md`). Some aspects of the workflow are automated with Snakemake ([`Snakefile`](Snakefile)). 
+Before starting, you should have already run the scripts `00` and `01` folders to produce mock baselines. Then, you will need to copy the `mock_data.npz` into this directory in a new `data` folder. For example, from within this 02 folder, run
+
+```shell
+$ mkdir data
+$ cp ../01-generate-mock-baselines/data/mock_data.npz data/
+```
+
+# Installation
+
+You can install necessary Python packages into your environment by
+```shell
+$ pip install -r requirements.txt
+```
+
+and then you can run the code by
+
+```shell 
+snakemake -c1 all
+```
+
+# Description of Contents
+
+This repository assumes that you will run all scripts from this `02` directory (the one containing `02-sgd/README.md`). Some aspects of the workflow are automated with Snakemake ([`Snakefile`](Snakefile)). 
 
 First, we recommend looking at [`src/load_data.py`](src/load_data.py) to see how mock visibilities $\mathcal{V}(u,v)$ are generated from the mock image and baselines.
 
 Then, we recommend looking at [`src/plot_baselines.py`](src/plot_baselines.py) and [`src/dirty_image.py`](src/dirty_image.py) to make diagnostic plots of the baseline and a dirty image of the data, to check that everything appears as you might expect.
 
+You can run these simple scripts using 
+
+```
+$ snakemake -c1 all
+```
+
+![baselines](analysis/baselines.png)
+
+![Dirty Beam and Image](analysis/dirty_image.png)
+
+# RML imaging workflow
+
 The RML imaging workflow is demonstrated in [`src/sgd.py`](src/sgd.py). We recommend looking through that file before reading the rest of this document. If you are new to PyTorch idioms, we recommend familiarizing yourself with the [PyTorch basics](https://mpol-dev.github.io/MPoL/background.html#pytorch) first. 
 
+The RML imaging workflow is not part of the Snakemake workflow, instead, one runs the script like
+
+```
+$ python src/sgd.py --epochs=5
+```
+
+Note this will just result in a short test. Run `python src/sgd.py --help` to see all available command line arguments, and see below for configurations that will result in better images.
+
+One can visualize the results using Tensorboard via
+
+```shell
+$ tensorboard --logdir runs
+```
+
 # Validation 
 Since this example uses mock data, we have the advantage of knowing the true sky image. This allows us to calculate a 'validation loss' between the synthesized image and the true sky.
 
@@ -61,5 +109,4 @@ Adding entropy regularization, and reducing learning rate slightly.
 python src/sgd.py --tensorboard-log-dir=runs/ent0 --load-checkpoint=checkpoints/2.pt --save-checkpoint=checkpoints/ent0.pt --lr 1e-1 --FWHM 0.05 --epochs=50 --lam-ent=1e-5
 ```
 
-
 Note that we could have started directly with the entropy regularization if we wished. The previous just demonstrates an exploratory workflow.

02-sgd/requirements.txt

@@ -1,3 +1,5 @@
 mpol
+tensorboard
 visread
-matplotlib
+matplotlib
+snakemake

02-sgd/src/sgd.py

@@ -257,6 +257,7 @@ def main():
         vis_data.uu, vis_data.vv, vis_data.weight, vis_data.data
     )
 
+    print("running on ", device)
     print("total vis", len(train_dataset))
 
     # set the batch sizes for the loaders