Simple scheduler for executing detailed command-line programs

• About
• Getting Started
• Usage

Agenda is a tool for scheduling command-line programs with too many arguments. It allows the user to configure many runs at once in a configuration file and run a single, simple command that will take care of the rest. This makes running tests, experiments and per-batch scripts a lot easier.

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Agenda is built using Python3. Install dependencies from src/requirements.txt:

pip3 install -r requirements.txt

Locally install agenda with pip so it can be used from anywhere in your machine:

pip3 install -e .

Automated tests are implemented with PyTest. After installing project requirements, run pytest from the src directory.

The goal of agenda is to be simple. All you need is a configuration JSON file and a goal in mind.

As an example, suppose you have a Python program that trains a neural network on one of a range of possible train datasets and then sends its accuracy on a test dataset to someone's email address. You would normally run it by writing python3 main.py --train TRAIN_DATASET --test TEST_DATASET --target_email EMAIL@PROVIDER.COM. This is okay if you're doing it once, but in many cases you'll want to run this command with many different values for train, test and target_email. You would have to type that out, run the first experiment, then type it out again with a different set of arguments, run the second experiment, and so it goes.

With agenda, your work is put into deciding what you'll be executing. The only command you run is actually pretty simple. You can configure these runs like so, in a JSON file:

{
  "required_keys": ["train", "test"],
  "log_header": "This is really easy!",
  "prefix": "python3 main.py",
  "cmds": [
    {
      "train": "train_dataset_1",
      "test": "test_dataset_1"
    },
    {
      "train": "train_dataset_1",
      "test": "test_dataset_2",
      "target_email": "email@provider.com"
    },
    {
      "env_variables": {
        "CUDA_VISIBLE_DEVICES": "5,7"
      },
      "train": "train_dataset_2",
      "test": "test_dataset_3"
    }
  ]
}

Let's go through that step by step.

• required_keys is optional; if it is provided, agenda will look for these arguments in every command and, if it isn't there, it will raise an error.
• log_header is also optional and lets you write some initial message to the log file.
• prefix is a common prefix in all of the commands.
• cmds is a list of all the commands that will be run. Each of its elements is a dictionary and each key in this dictionary is a command-line argument to be run, except for...
• env_variables (optional) contains environment variables to be set before the command runs.

Finally, you can easily run all of these experiments with a single command, like so:

python3 agenda.py --config config.json --output output.txt

This is equivalent to these 3 commands:

python3 main.py --train train_dataset_1 --test test_dataset_1
python3 main.py --train train_dataset_1 --test test_dataset_2 --target_email email@provider.com
CUDA_VISIBLE_DEVICES=5,7 python3 main.py --train train_dataset_2 --test test_dataset_3