added octorotor and commonml submodules

[submodule "commonml"]

	path = commonml

	url = git@git.isis.vanderbilt.edu:ahmedi/commonml.git

[submodule "systems/octorotor"]

	path = systems/octorotor

	url = git@git.isis.vanderbilt.edu:ahmedi/gym-octorotor.git

Subproject commit b702383cd3b00173f9db5990b6d5ae451d24a371

import torch

from torch.optim.lr_scheduler import CosineAnnealingLR

from higher import innerloop_ctx

import numpy as np

from tqdm.auto import tqdm, trange

import matplotlib.pyplot as plt

from sklearn.neural_network import MLPRegressor

from scipy.spatial.distance import jensenshannon, cosine, euclidean

from utils import copy_mlp_regressor, copy_tensor, higher_dummy_context, get_gradients

from ppo import PPO, Memory, DEVICE, returns

from utils import copy_mlp_regressor, copy_tensor, get_gradients

from ppo import PPO, Memory, DEVICE

    vals = []

    for params in library:

        agent.policy.load_state_dict(params)

        # TODO: logp is p of taking action from state,

        # shouldnt we multiply logp of (state-1, action-1) with the value

        # of state to get the weighed value of each state in sequence times

        # the likelihood of that state?

        logp, _, _ = agent.policy.evaluate(states, actions)

        p = torch.exp(logp)

        vals.append(torch.sum(p * returns).item())

        # multiply probability of action reaching that state with value of state

        vals.append(torch.sum(p[:-1] * returns[1:]).item())

    agent.policy.load_state_dict(og_params)

    return np.argsort(vals)[::-1], np.asarray(vals)

import numpy as np

from tqdm.auto import trange

from utils import higher_dummy_context

# pylint: disable=no-member

DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")