Adding, updating, and fixing docs. (#53)

* adding docs

* Fix, update, and document mini-imagenet dataset.

* MiniImagenetDataset -> MiniImagenet for convention.

* Add omniglot docs.

* Add docs for RandomDiscreteRotations.

* Add docs for OmniglotFC.

* Add docs for OmniglotCNN and MiniImagenetCNN.

* Fix docs for particles env.

* Fix linting.

docs/pydocmd.yml

@@ -33,8 +33,10 @@ generate:
  - learn2learn.vision.models
  - learn2learn.vision.models.OmniglotFC
  - learn2learn.vision.models.OmniglotCNN
+ - learn2learn.vision.models.MiniImagenetCNN
  - learn2learn.vision.datasets
  - learn2learn.vision.datasets.FullOmniglot
+ - learn2learn.vision.datasets.MiniImagenet
  - learn2learn.vision.transforms
  - learn2learn.vision.transforms.RandomDiscreteRotation
  - docs/learn2learn.text.md:
@@ -53,7 +55,6 @@ pages:
  - learn2learn.algorithms: docs/learn2learn.algorithms.md
  - learn2learn.data: docs/learn2learn.data.md
  - learn2learn.gym: docs/learn2learn.gym.md
- - learn2learn.text: docs/learn2learn.text.md
  - learn2learn.vision: docs/learn2learn.vision.md
  - Examples: https://github.com/learnables/learn2learn/tree/master/examples
  - GitHub: https://github.com/learnables/learn2learn/

examples/vision/maml_miniimagenet.py

@@ -47,27 +47,18 @@ def main(
  cuda=True,
  seed=42,
 ):
- MI_PATH = '~/Dropbox/Temporary/mini-imagenet-l2l/miniimagenet/resized/'
-
  random.seed(seed)
  np.random.seed(seed)
  th.manual_seed(seed)
  device = th.device('cpu')
- if cuda:
+ if cuda and th.cuda.device_count():
  th.cuda.manual_seed(seed)
  device = th.device('cuda')
 
- # Create Dataset
- transform = transforms.Compose([
- transforms.ToTensor(),
- lambda x: x.float() / 255.,
- ])
- train_images_path = os.path.join(MI_PATH, 'train')
- valid_images_path = os.path.join(MI_PATH, 'val')
- test_images_path = os.path.join(MI_PATH, 'test')
- train_dataset = ImageFolder(train_images_path, transform)
- valid_dataset = ImageFolder(valid_images_path, transform)
- test_dataset = ImageFolder(test_images_path, transform)
+ # Create Datasets
+ train_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='train')
+ valid_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='validation')
+ test_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='test')
  train_dataset = l2l.data.MetaDataset(train_dataset)
  valid_dataset = l2l.data.MetaDataset(valid_dataset)
  test_dataset = l2l.data.MetaDataset(test_dataset)

learn2learn/algorithms/maml.py

@@ -119,7 +119,7 @@ def adapt(self, loss, first_order=None):
  * **loss** (Tensor) - Loss to minimize upon update.
  * **first_order** (bool, *optional*, default=None) - Whether to use first- or
  second-order updates. Defaults to self.first_order.
- 
+
  """
  if first_order is None:
  first_order = self.first_order
@@ -145,7 +145,7 @@ def clone(self, first_order=None):
 
  * **first_order** (bool, *optional*, default=None) - Whether the clone uses first-
  or second-order updates. Defaults to self.first_order.
- 
+
  """
  if first_order is None:
  first_order = self.first_order

learn2learn/data/task_generator.py

@@ -23,8 +23,8 @@ def __getitem__(self, idx):
 
 
 class MetaDataset(Dataset):
- """ 
- 
+ """
+
  **Descritpion**
 
  It wraps a torch dataset by creating a map of target to indices.
@@ -37,7 +37,7 @@ class MetaDataset(Dataset):
  #TODO : Add example for wrapping a non standard l2l dataset
 
  **Arguments**
- 
+
  * **dataset** (Dataset) - A torch dataset.
 
  **Example**
@@ -191,7 +191,7 @@ def sample(self, shots=None, task=None):
  **task** (list, *optional*, default=None) - List of labels you want to sample from.
 
  **Returns**
- 
+
  * Dataset - Containing the sampled task.
 
  """

learn2learn/gym/envs/meta_env.py

@@ -47,7 +47,7 @@ def sample_tasks(self, num_tasks):
 
  **Returns**
 
- tasks ([dict]) - returns a list of num_tasks tasks. Tasks are 
+ tasks ([dict]) - returns a list of num_tasks tasks. Tasks are
  dictionaries of task specific parameters. A
  minimal example for num_tasks = 1 is [{'goal': value}].
  """
@@ -57,7 +57,7 @@ def set_task(self, task):
  """
  **Description**
 
- Sets the task specific parameters defined in task. 
+ Sets the task specific parameters defined in task.
 
  **Arguments**
 

learn2learn/gym/envs/mujoco/ant_direction.py

@@ -9,18 +9,18 @@
 
 class AntDirectionEnv(MetaEnv, MujocoEnv, gym.utils.EzPickle):
  """
- [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/ant_direction.py) 
+ [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/ant_direction.py)
 
  **Description**
 
  This environment requires the Ant to learn to run in a random direction in the
  XY plane. At each time step the ant receives a signal composed of a
  control cost and a reward equal to its average velocity in the direction
  of the plane. The tasks are 2d-arrays sampled uniformly along the unit circle.
- The target direction is indicated by the vector from the origin to the sampled point. 
+ The target direction is indicated by the vector from the origin to the sampled point.
  The velocity is calculated as the distance (in the target direction) of the ant's torso
  position before and after taking the specified action divided by a small value dt.
- As noted in [1], a small positive bonus is added to the reward to stop the ant from 
+ As noted in [1], a small positive bonus is added to the reward to stop the ant from
  prematurely ending the episode.
 
  **Credit**
@@ -39,7 +39,7 @@ def __init__(self, task=None):
  MujocoEnv.__init__(self, 'ant.xml', 5)
  gym.utils.EzPickle.__init__(self)
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def set_task(self, task):
  MetaEnv.set_task(self, task)
  self.goal_direction = task['direction']
@@ -50,7 +50,7 @@ def sample_tasks(self, num_tasks):
  tasks = [{'direction': direction} for direction in directions]
  return tasks
 
- # -------- Mujoco Methods -------- 
+ # -------- Mujoco Methods --------
  def _get_obs(self):
  return np.concatenate([
  self.sim.data.qpos.flat[2:],
@@ -71,7 +71,7 @@ def reset_model(self):
  qvel = self.init_qvel + self.np_random.randn(self.model.nv) * .1
  self.set_state(qpos, qvel)
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def step(self, action):
  posbefore = np.copy(self.get_body_com("torso")[:2])
  self.do_simulation(action, self.frame_skip)

learn2learn/gym/envs/mujoco/ant_forward_backward.py

@@ -9,18 +9,18 @@
 
 class AntForwardBackwardEnv(MetaEnv, MujocoEnv, gym.utils.EzPickle):
  """
- [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/ant_forward_backward.py) 
+ [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/ant_forward_backward.py)
 
  **Description**
 
- This environment requires the ant to learn to run forward or backward. 
+ This environment requires the ant to learn to run forward or backward.
  At each time step the ant receives a signal composed of a
  control cost and a reward equal to its average velocity in the direction
  of the plane. The tasks are Bernoulli samples on {-1, 1} with probability 0.5, where -1 indicates the ant should
  move backward and +1 indicates the ant should move forward.
  The velocity is calculated as the distance (in the direction of the plane) of the ant's torso
  position before and after taking the specified action divided by a small value dt.
- As noted in [1], a small positive bonus is added to the reward to stop the ant from 
+ As noted in [1], a small positive bonus is added to the reward to stop the ant from
  prematurely ending the episode.
 
  **Credit**
@@ -39,7 +39,7 @@ def __init__(self, task=None):
  MujocoEnv.__init__(self, 'ant.xml', 5)
  gym.utils.EzPickle.__init__(self)
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def set_task(self, task):
  MetaEnv.set_task(self, task)
  self.goal_direction = task['direction']
@@ -49,7 +49,7 @@ def sample_tasks(self, num_tasks):
  tasks = [{'direction': direction} for direction in directions]
  return tasks
 
- # -------- Mujoco Methods -------- 
+ # -------- Mujoco Methods --------
  def _get_obs(self):
  return np.concatenate([
  self.sim.data.qpos.flat[2:],
@@ -70,7 +70,7 @@ def reset_model(self):
  qvel = self.init_qvel + self.np_random.randn(self.model.nv) * .1
  self.set_state(qpos, qvel)
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def step(self, action):
  xposbefore = np.copy(self.get_body_com("torso")[0])
  self.do_simulation(action, self.frame_skip)

learn2learn/gym/envs/mujoco/halfcheetah_forward_backward.py

@@ -10,10 +10,10 @@
 class HalfCheetahForwardBackwardEnv(MetaEnv, MujocoEnv, gym.utils.EzPickle):
  """
  [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/halfcheetah_forward_backward.py)
- 
+
  **Description**
 
- This environment requires the half-cheetah to learn to run forward or backward. 
+ This environment requires the half-cheetah to learn to run forward or backward.
  At each time step the half-cheetah receives a signal composed of a
  control cost and a reward equal to its average velocity in the direction
  of the plane. The tasks are Bernoulli samples on {-1, 1} with probability 0.5, where -1 indicates the half-cheetah should
@@ -37,7 +37,7 @@ def __init__(self, task=None):
  MujocoEnv.__init__(self, 'half_cheetah.xml', 5)
  gym.utils.EzPickle.__init__(self)
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def set_task(self, task):
  MetaEnv.set_task(self, task)
  self.goal_direction = task['direction']
@@ -47,7 +47,7 @@ def sample_tasks(self, num_tasks):
  tasks = [{'direction': direction} for direction in directions]
  return tasks
 
- # -------- Mujoco Methods -------- 
+ # -------- Mujoco Methods --------
  def _get_obs(self):
  return np.concatenate([
  self.sim.data.qpos.flat[1:],
@@ -69,7 +69,7 @@ def reset_model(self):
  self.set_state(qpos, qvel)
  return self._get_obs()
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def step(self, action):
  xposbefore = self.sim.data.qpos[0]
  self.do_simulation(action, self.frame_skip)

learn2learn/gym/envs/mujoco/humanoid_direction.py

@@ -23,10 +23,10 @@ class HumanoidDirectionEnv(MetaEnv, MujocoEnv, gym.utils.EzPickle):
  XY plane. At each time step the humanoid receives a signal composed of a
  control cost and a reward equal to its average velocity in the target direction.
  The tasks are 2d-arrays sampled uniformly along the unit circle.
- The target direction is indicated by the vector from the origin to the sampled point. 
+ The target direction is indicated by the vector from the origin to the sampled point.
  The velocity is calculated as the distance (in the target direction) of the humanoid's torso
  position before and after taking the specified action divided by a small value dt.
- A small positive bonus is added to the reward to stop the humanoid from 
+ A small positive bonus is added to the reward to stop the humanoid from
  prematurely ending the episode.
 
  **Credit**
@@ -45,7 +45,7 @@ def __init__(self, task=None):
  MujocoEnv.__init__(self, 'humanoid.xml', 5)
  gym.utils.EzPickle.__init__(self)
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def set_task(self, task):
  MetaEnv.set_task(self, task)
  self.goal_direction = task['direction']
@@ -56,7 +56,7 @@ def sample_tasks(self, num_tasks):
  tasks = [{'direction': direction} for direction in directions]
  return tasks
 
- # -------- Mujoco Methods -------- 
+ # -------- Mujoco Methods --------
  def _get_obs(self):
  data = self.sim.data
  return np.concatenate([data.qpos.flat[2:],
@@ -79,7 +79,7 @@ def reset_model(self):
  )
  return self._get_obs()
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def step(self, action):
  pos_before = np.copy(mass_center(self.model, self.sim)[:2])
  self.do_simulation(action, self.frame_skip)

learn2learn/gym/envs/mujoco/humanoid_forward_backward.py

@@ -16,10 +16,10 @@ def mass_center(model, sim):
 class HumanoidForwardBackwardEnv(MetaEnv, MujocoEnv, gym.utils.EzPickle):
  """
  [[Source]](https://github.com/learnables/learn2learn/blob/master/learn2learn/gym/envs/mujoco/humanoid_forward_backward.py)
- 
+
  **Description**
 
- This environment requires the humanoid to learn to run forward or backward. 
+ This environment requires the humanoid to learn to run forward or backward.
  At each time step the humanoid receives a signal composed of a
  control cost and a reward equal to its average velocity in the target direction.
  The tasks are Bernoulli samples on {-1, 1} with probability 0.5, where -1 indicates the humanoid should
@@ -43,7 +43,7 @@ def __init__(self, task=None):
  MujocoEnv.__init__(self, 'humanoid.xml', 5)
  gym.utils.EzPickle.__init__(self)
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def set_task(self, task):
  MetaEnv.set_task(self, task)
  self.goal_direction = task['direction']
@@ -53,7 +53,7 @@ def sample_tasks(self, num_tasks):
  tasks = [{'direction': direction} for direction in directions]
  return tasks
 
- # -------- Mujoco Methods -------- 
+ # -------- Mujoco Methods --------
  def _get_obs(self):
  data = self.sim.data
  return np.concatenate([data.qpos.flat[2:],
@@ -76,7 +76,7 @@ def reset_model(self):
  )
  return self._get_obs()
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def step(self, action):
  pos_before = mass_center(self.model, self.sim)[0]
  self.do_simulation(action, self.frame_skip)

learn2learn/gym/envs/particles/particles_2d.py

@@ -14,19 +14,14 @@ class Particles2DEnv(MetaEnv):
  **Description**
 
  Each task is defined by the location of the goal. A point mass
- receives a directional force and moves accordingly 
+ receives a directional force and moves accordingly
  (clipped in [-0.1,0.1]). The reward is equal to the negative
  distance from the goal.
 
  **Credit**
 
  Adapted from Jonas Rothfuss' implementation.
 
- **Arguments**
-
- task (dict, *optional*) - Environment can optionally be initialized
- with a task. If no task is provided, one is randomly sampled.
-
  """
 
  def __init__(self, task=None):
@@ -38,10 +33,10 @@ def __init__(self, task=None):
  shape=(2,), dtype=np.float32)
  self.reset()
 
- # -------- MetaEnv Methods -------- 
+ # -------- MetaEnv Methods --------
  def sample_tasks(self, num_tasks):
  """
- Tasks correspond to a goal point chosen uniformly at random. 
+ Tasks correspond to a goal point chosen uniformly at random.
  """
  goals = self.np_random.uniform(-0.5, 0.5, size=(num_tasks, 2))
  tasks = [{'goal': goal} for goal in goals]
@@ -51,7 +46,7 @@ def set_task(self, task):
  self._task = task
  self._goal = task['goal']
 
- # -------- Gym Methods -------- 
+ # -------- Gym Methods --------
  def seed(self, seed=None):
  self.np_random, seed = seeding.np_random(seed)
  return [seed]
@@ -66,7 +61,7 @@ def reset(self, env=True):
  def step(self, action):
  """
  **Description**
- 
+
  Given an action, clips the action to be in the
  appropriate range and moves the point mass position
  according to the action.
@@ -76,7 +71,7 @@ def step(self, action):
  action (2-element array) - Array specifying the magnitude
  and direction of the forces to be applied in the x and y
  planes.
- 
+
  **Returns**
 
  *state, reward, done, task*