Berkeley CS 294 Homework 1

run_dagger.py

#!/usr/bin/env python

"""
Code to load an expert policy and generate roll-out data for behavioral cloning.
Example usage:
    python run_expert.py experts/Humanoid-v1.pkl Humanoid-v1 --render \
            --num_rollouts 20

Author of this script and included expert policies: Jonathan Ho ([email protected])
"""

import pickle
import tensorflow as tf
import numpy as np
import numpy.random as RNG
import tf_util
import gym
import load_policy
import tensorflow.contrib.keras as K
import h5py

def main():
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument('expert_policy_file', type=str)
    parser.add_argument('envname', type=str)
    parser.add_argument('input', type=str)
    parser.add_argument('--render', action='store_true')
    parser.add_argument("--max_timesteps", type=int)
    parser.add_argument("--datacap", type=int, default=100000)
    parser.add_argument('--num_rollouts', type=int, default=20,
                        help='Number of expert roll outs')
    args = parser.parse_args()

    print('loading and building student policy')
    data = h5py.File(args.input)
    x = np.copy(data['observations'][:args.datacap])
    a = np.copy(data['actions'][:args.datacap])
    data.close()
    input_dim = x.shape[1]
    output_dim = a.shape[2]
    model = K.models.Sequential()
    model.add(K.layers.Lambda(lambda x: x / 10, input_shape=(input_dim,)))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=output_dim))
    model.compile(loss='mean_squared_error',
                  optimizer='rmsprop')
    def policy_fn(obs):
        return model.predict(obs, batch_size=obs.shape[0])
    policy_fn_expert = load_policy.load_policy(args.expert_policy_file)

    print('loaded and built')

    with tf.Session():
        tf_util.initialize()

        env = gym.make(args.envname)
        max_steps = args.max_timesteps or env.spec.timestep_limit

        returns = []
        observations = []
        actions = []
        for i in range(args.num_rollouts):
            current_obs = []
            current_acts = []
            print('iter', i)
            obs = env.reset()
            done = False
            totalr = 0.
            steps = 0
            model.fit(x, a[:, 0], batch_size=200, epochs=5, shuffle='batch')
            while not done:
                action = policy_fn(obs[None,:])
                current_obs.append(obs)
                current_acts.append(action)
                obs, r, done, _ = env.step(action)
                totalr += r
                steps += 1
                if args.render:
                    env.render()
                if steps % 100 == 0: print("%i/%i"%(steps, max_steps))
                if steps >= max_steps:
                    break
            returns.append(totalr)

            idx = RNG.choice(args.datacap, len(current_obs))
            x[idx] = current_obs
            a[idx, 0] = policy_fn_expert(current_obs)

            observations.extend(current_obs)
            actions.extend(current_acts)

        print('returns', returns)
        print('mean return', np.mean(returns))
        print('std of return', np.std(returns))

        expert_data = {'observations': np.array(observations),
                       'actions': np.array(actions)}

if __name__ == '__main__':
    main()

run_student.py

#!/usr/bin/env python

"""
Code to load an expert policy and generate roll-out data for behavioral cloning.
Example usage:
    python run_expert.py experts/Humanoid-v1.pkl Humanoid-v1 --render \
            --num_rollouts 20

Author of this script and included expert policies: Jonathan Ho ([email protected])
"""

import pickle
import tensorflow as tf
import numpy as np
import tf_util
import gym
import load_policy
import tensorflow.contrib.keras as K
import h5py

def main():
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument('expert_policy_file', type=str)
    parser.add_argument('envname', type=str)
    parser.add_argument('input', type=str)
    parser.add_argument('valid', type=str)
    parser.add_argument('--render', action='store_true')
    parser.add_argument("--max_timesteps", type=int)
    parser.add_argument('--num_rollouts', type=int, default=20,
                        help='Number of expert roll outs')
    args = parser.parse_args()

    print('loading and building student policy')
    data = h5py.File(args.input)
    valid = h5py.File(args.valid)
    x = data['observations'][:100000]
    a = data['actions'][:100000]
    xv = valid['observations']
    av = valid['actions']
    input_dim = x.shape[1]
    output_dim = a.shape[2]
    model = K.models.Sequential()
    model.add(K.layers.Lambda(lambda x: x / 10, input_shape=(input_dim,)))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=100))
    model.add(K.layers.Activation('tanh'))
    model.add(K.layers.Dense(units=output_dim))
    model.compile(loss='mean_squared_error',
                  optimizer='rmsprop')
    def policy_fn(obs):
        return model.predict(obs, batch_size=obs.shape[0])
    print('loaded and built')

    with tf.Session():
        tf_util.initialize()

        import gym
        env = gym.make(args.envname)
        max_steps = args.max_timesteps or env.spec.timestep_limit

        returns = []
        observations = []
        actions = []
        model.fit(x, a[:, 0], batch_size=200, epochs=50, shuffle='batch',
                  callbacks=[K.callbacks.EarlyStopping(patience=10)],
                  validation_data=(xv, av[:, 0]))
        for i in range(args.num_rollouts):
            print('iter', i)
            obs = env.reset()
            done = False
            totalr = 0.
            steps = 0
            while not done:
                action = policy_fn(obs[None,:])
                observations.append(obs)
                actions.append(action)
                obs, r, done, _ = env.step(action)
                totalr += r
                steps += 1
                if args.render:
                    env.render()
                if steps % 100 == 0: print("%i/%i"%(steps, max_steps))
                if steps >= max_steps:
                    break
            returns.append(totalr)

        print('returns', returns)
        print('mean return', np.mean(returns))
        print('std of return', np.std(returns))

        expert_data = {'observations': np.array(observations),
                       'actions': np.array(actions)}

if __name__ == '__main__':
    main()