Properly Initialize Repo + Code Health

.github/workflows/python-test-coverage.yml

@@ -35,4 +35,4 @@ jobs:
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} 
       run: |
-        coveralls
+        coveralls --service=github

auto_trainer/test_trainer.py

@@ -0,0 +1,137 @@
+import unittest
+
+from auto_trainer import trainer
+from unittest import mock
+
+import importlib
+import sys
+
+
+class TestAutoTrainer(unittest.TestCase):
+  def test_synced_config_no_wandb(self):
+    param = {'test_key': 'test_value'}
+
+    trainer._WANDB = False
+    self.assertFalse(trainer._WANDB)
+
+    config, run = trainer.get_synced_config(param, ['bunch', 'of', 'tags'])
+
+    self.assertDictEqual(config, param)
+    self.assertIsNone(run)
+
+  def test_synced_config_wandb(self):
+    mock_wandb = mock.MagicMock()
+
+    param = {'test_key': 'test_value'}
+    tags = ['bunch', 'of', 'tags']
+
+    mock_run = mock.MagicMock(config=param)
+    mock_wandb.init = mock.MagicMock()
+    mock_wandb.init.return_value = mock_run
+
+    if 'wandb' in sys.modules: 
+      import wandb
+      del wandb
+    with mock.patch.dict('sys.modules', {'wandb': mock_wandb}):
+      importlib.reload(trainer)
+
+      trainer._WANDB = True
+      self.assertTrue(trainer._WANDB)
+      config, run = trainer.get_synced_config(param, tags)
+
+    mock_wandb.init.assert_called_once()
+
+    _, kwargs = mock_wandb.init.call_args
+    self.assertDictEqual(kwargs['config'], param)
+    self.assertListEqual(kwargs['tags'], tags)
+    self.assertDictEqual(config, param)
+    self.assertEqual(run, mock_run)
+
+  def test_trainer_no_wandb(self):
+    trainer._WANDB = False
+    self.assertFalse(trainer._WANDB)
+
+    algo = 'test_ago'
+    policy = 'test_policy'
+    episodes = 69
+
+    parameters = mock.MagicMock(algorithm=algo, policy=policy, 
+                                episodes=episodes)
+    fake_env = mock.MagicMock()
+
+    mock_learn = mock.MagicMock()
+    mock_save = mock.MagicMock()
+    mock_model = mock.MagicMock()
+    mock_model.learn = mock_learn
+    mock_model.save = mock_save
+
+    mock_model_cls = mock.MagicMock()
+    mock_model_cls.return_value = mock_model
+
+    with mock.patch.dict(trainer.SUPPORTED_ALGORITHMS, 
+                         {algo: mock_model_cls}, clear=True):
+      model, config, run = trainer.train(fake_env, parameters, None)
+
+      self.assertEqual(model, mock_model)
+      self.assertEqual(config, parameters)
+      self.assertIsNone(run)
+
+      mock_model_cls.assert_called_once()
+      args, _ = mock_model_cls.call_args
+      self.assertTupleEqual(args, (policy, fake_env))
+
+      mock_learn.assert_called_once()
+      mock_learn_args, _ = mock_learn.call_args
+      self.assertTupleEqual(mock_learn_args, (episodes, ))
+
+      mock_save.assert_called_once()
+      mock_save_args, _ = mock_save.call_args
+
+      # If not using wandb, should save in 2-digit reps of 
+      # MonthDayHourMinSec; henceforth, the length of the run name should be 10 
+      # digits long
+      self.assertEqual(len(mock_save_args[0]), 10)
+
+  def test_trainer_wandb(self):
+    algo = 'test_ago'
+    policy = 'test_policy'
+    episodes = 69
+
+    parameters = mock.MagicMock(algorithm=algo, policy=policy, 
+                                episodes=episodes)
+    mock_env = mock.MagicMock()
+    mock_run = mock.MagicMock(dir='test_dir')
+
+    mock_learn = mock.MagicMock()
+    mock_save = mock.MagicMock()
+    mock_model = mock.MagicMock()
+    mock_model.learn = mock_learn
+    mock_model.save = mock_save
+
+    mock_model_cls = mock.MagicMock()
+    mock_model_cls.return_value = mock_model
+
+    mock_wandb = mock.MagicMock()
+    if 'wandb' in sys.modules: 
+      import wandb
+      del wandb
+    with mock.patch.dict('sys.modules', {'wandb': mock_wandb}):
+      importlib.reload(trainer)
+
+      trainer._WANDB = True
+      self.assertTrue(trainer._WANDB)
+
+      with mock.patch.dict(trainer.SUPPORTED_ALGORITHMS, 
+                          {algo: mock_model_cls}, clear=True):
+        model, config, run = trainer.train(mock_env, parameters, None, 
+                                          run=mock_run)
+
+    _, kwargs = mock_model_cls.call_args
+    self.assertEqual(kwargs['tensorboard_log'], mock_run.dir)
+
+    args, _ = mock_save.call_args
+    self.assertEqual(args[0], '{}/model'.format(mock_run.dir))
+
+
+if __name__ == '__main__':
+  pass

auto_trainer/trainer.py

@@ -1,7 +1,11 @@
+from typing import List, Text
+
 from datetime import datetime
-import stable_baselines
+
+import gym
 import logging
 import os
+import stable_baselines
 
 
 PROJECT_NAME = 'solo-rl-experiments'
@@ -21,7 +25,22 @@
   _WANDB = False
 
 
-def get_synced_config(parameters, tags):
+def get_synced_config(parameters, tags: List[Text]):
+  """Sync the config with wandb (if necessary) and return the new config
+
+  Args:
+    parameters (Any W&B supported type): The current hyperparameters to use. If
+      W&B is enabled and is actively making a sweep, these hyperparameters will 
+      get updated to W&B's sweep. This can be any type supported by W&B, 
+      including Dicts and argparse.Namespace objects.
+    tags (List[Text]): Tags that describe the run. Note that this is basically
+      useless if W&B is disabled.
+
+  Returns:
+    The (hyperparameter config, W&B run object). Obviously, if W&B is disabled,
+    then the run object will be None and the hyperparameter config will be what
+    was passed in. 
+  """
   if not _WANDB:
     return parameters, None
 
@@ -37,7 +56,27 @@ def get_synced_config(parameters, tags):
   return config, run
 
 
-def train(env, parameters, tags, full_logging=True, log_freq=100, run=None):
+def train(env: gym.Env, parameters, tags: List[Text], 
+          full_logging: bool = False, log_freq: int = 100, run = None):
+  """Train a model.
+
+  Args:
+    env (gym.Env): Gym environment to train on.
+    parameters (Any W&B supported type): The hyperparameters to train with.
+      Refer to W&B for all of the support types.
+    tags (List[Text]): List of tags that describe this run. Doesn't do anything
+      if `run` is not None.
+    full_logging (bool, optional): Whether or not to log *everything*. Can fill
+      up space quick. Defaults to True.
+    log_freq (int, optional): How many steps to write the logs. Defaults to 100.
+    run ([wandb.Run], optional): A current W&B run. Use this if you want to
+      reuse a current run, i.e. train a model, do things to it, and continue
+      training it. If this is None, a new run will be created via
+      `get_synced_config`. Defaults to None.
+
+  Returns:
+    [type]: [description]
+  """
   if run:
     config = parameters
   else: 
@@ -50,7 +89,7 @@ def train(env, parameters, tags, full_logging=True, log_freq=100, run=None):
                     verbose=1)
 
   if _WANDB: wandb.tensorboard.monkeypatch._notify_tensorboard_logdir(
-      os.path.join(run.dir, '{}_1'.format(_DEFAULT_RUN_NAME)))
+    os.path.join(run.dir, '{}_1'.format(_DEFAULT_RUN_NAME)))
 
   model.learn(config.episodes, tb_log_name=_DEFAULT_RUN_NAME,
               log_interval=log_freq)

experiments/notebooks/solo8v2vanilla-ppo2.ipynb

@@ -435,7 +435,7 @@
     "extension": ".py",
     "format_name": "percent",
     "format_version": "1.3",
-    "jupytext_version": "1.7.1"
+    "jupytext_version": "1.9.1"
    }
   },
   "kernelspec": {

experiments/train/autotrainer-sb-solo-demo.py

@@ -5,7 +5,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.7.1
+#       jupytext_version: 1.9.1
 #   kernelspec:
 #     display_name: Python 3
 #     language: python

experiments/train/solo8v2vanilla-ppo2.py

@@ -5,7 +5,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.7.1
+#       jupytext_version: 1.9.1
 #   kernelspec:
 #     display_name: Python 3
 #     language: python

setup.py

@@ -9,6 +9,7 @@
         'stable-baselines', 
         'numpy<1.19.0,>=1.16.0',
         'jupytext',
+        'gym'
       ], 
       extras_require={
         'cpu': ['tensorflow>=1.15.0,<2'],