Random Designer Processing Improvements

[karldd]

This PR adds several improvements for data processing to the random designer:

• A timer to recycle Fusion after 5 minutes on any one design generation
• Ability to run multiple random designers in parallel on the same machine
• Pass arguments to the random designer
• Only select from training set sketches

Usage

usage: random_designer.py [-h] [--input INPUT] [--split SPLIT]
                          [--output OUTPUT] [--machine_id MACHINE_ID]
                          [--host HOST] [--port PORT]

optional arguments:
  -h, --help            show this help message and exit
  --input INPUT         File or folder containing the dataset [default: d7]
  --split SPLIT         Train/test split file from which to select train
                        sketches only [default: train_test.json]
  --output OUTPUT       Folder to save the output to [default:
                        generated_design]
  --machine_id MACHINE_ID
                        Machine id used in file names [default: 2]
  --host HOST           Host name to launch Fusion Gym on [default: 127.0.0.1]
  --port PORT           Port number to launch Fusion Gym on [default: 8080]

For example, to run multiple at the same time you would do something like this in one terminal:

 python random_designer.py --input path/to/d7 --split path/to/train_test.json --output path/to/output --machine_id 8 --port 8083

Then in another terminal:

 python random_designer.py --input path/to/d7 --split path/to/train_test.json --output path/to/output --machine_id 7 --port 8084

Note that the code now assumes Fusion is not running, and the gym server is set to run on startup. In general you can run up to ~8 instances on the same computer depending on its specs.