Python load_params示例

示例#1

    def setUp(self):
        """setUp"""
        self.params = load_params("params/test_params.json")

        # environment without normalization:
        self.params["env"]["normalize"] = False
        self.env = gym.make(self.params["env"]["name"], params=self.params)
        # environment with normaluzation:
        self.params["env"]["normalize"] = True
        self.env_norm = gym.make(self.params["env"]["name"],
                                 params=self.params)

示例#2

    def setUp(self):
        """setup"""
        self.params = load_params('params/test_params.json')

        self.env = gym.make(self.params["env"]["name"], params=self.params)
        self.test_env = gym.make(self.params["env"]["name"],
                                 params=self.params)

        self.policy = DDPG(env=self.env, params=self.params)

        self.cae = CAE(pooling='max',
                       latent_dim=16,
                       input_shape=(32, 32),
                       conv_filters=[4, 8, 16])
        self.cae.build(input_shape=(1, 32, 32, 1))
        self.cae.load_weights(filepath='../models/cae/model_num_5_size_8.h5')

示例#3

import sys
import glob
import shutil
import numpy as np
import tensorflow as tf

import gym
import gym_pointrobo

from hwr.agents.pointrobo_ddpg import DDPG
from hwr.cae.cae import CAE
from hwr.training.pointrobot_trainer import PointrobotTrainer
from hwr.utils import load_params, get_random_params, export_params

# loading params:
params = load_params('params/hyperparam_tuning_params.json')

# deleting the previous runs logs:
logdir_files = glob.glob(os.path.join('results', 'hyperparam_tuning'))
for f in logdir_files:
    if os.path.isdir(f):
        shutil.rmtree(f)
    else:
        os.remove(f)

for run in range(params["hyper_tuning"]["num_of_runs"]):

    # getting random hyperparams according to the ranges and placing them into params.
    params = get_random_params(params)

    # setting up logdir for the current hyperparams:

示例#4

文件： benchmark_trainings.py 项目： KatharinaHermann/tum-Advanced-DL-for-robotics-RL

import numpy as np
import tensorflow as tf
import glob
import json
import shutil

import gym
import gym_pointrobo

from hwr.agents.pointrobo_ddpg import DDPG
from hwr.cae.cae import CAE
from hwr.training.pointrobot_trainer import PointrobotTrainer
from hwr.utils import load_params, set_up_benchmark_params

# loading the params:
params = load_params('params/benchmark_trainings_empty.json')
benchmark_keys = params["benchmark"].keys()

# deleting the previous runs logs:
logdir_files = glob.glob(os.path.join(params["trainer"]["logdir"], "*"))
for f in logdir_files:
    if os.path.isdir(f):
        shutil.rmtree(f)
    else:
        os.remove(f)

for key in benchmark_keys:
    # loading original params:
    params = load_params('params/benchmark_trainings_empty.json')

    # setting up training run:

示例#5

文件： benchmark_evaluations.py 项目： KatharinaHermann/tum-Advanced-DL-for-robotics-RL

import numpy as np
import tensorflow as tf
import glob
import json
import shutil

import gym
import gym_pointrobo

from hwr.agents.pointrobo_ddpg import DDPG
from hwr.cae.cae import CAE
from hwr.training.pointrobot_trainer import PointrobotTrainer
from hwr.utils import load_params, set_up_benchmark_params

# loading the params:
params = load_params('params/benchmark_evaluations.json')
benchmark_keys = params["benchmark"].keys()

for key in benchmark_keys:
    print('------ Evaluation {} -------'.format(key))
    # loading original params:
    params = load_params('params/benchmark_evaluations.json')

    # setting up training run:
    params = set_up_benchmark_params(params, key)
    params["trainer"]["logdir"] = os.path.join(params["trainer"]["logdir"],
                                               key)

    #Initialize the environment
    env = gym.make(
        params["env"]["name"],

示例#6

        num_of_cases = 100
        for _ in range(num_of_cases):
            _, goal_pos_norm, agent_pos_norm = self.env_norm.reset()

            # assertions:
            self.assertTrue(
                np.isclose(rescale(goal_pos_norm, self.env_norm.pos_bounds),
                           self.env_norm.goal_pos,
                           atol=1e-6).all())
            self.assertTrue(
                np.isclose(rescale(agent_pos_norm, self.env_norm.pos_bounds),
                           self.env_norm.agent_pos,
                           atol=1e-6).all())


if __name__ == '__main__':

    params = load_params("params/test_params.json")
    params["env"]["grid_size"] = 20
    params["env"]["goal_reward"] = -0.01
    params["env"]["collision_reward"] = -1
    params["env"]["step_reward"] = -0.01
    params["env"]["max_goal_dist"] = 5

    test_pointrobot_gym_goal(params)
    test_pointrobot_gym_obstacle(params)
    test_pointrobot_gym_boundaries(params)
    print('All tests have run successfully!')

    unittest.main()

示例#7

import os
import sys
import numpy as np
import tensorflow as tf
import glob

import gym
import gym_pointrobo

from hwr.agents.pointrobo_ddpg import DDPG
from hwr.cae.cae import CAE
from hwr.training.pointrobot_trainer import PointrobotTrainer
from hwr.utils import load_params

# loading params:
params = load_params('params/pointrobot_training_params.json')

if params["trainer"]["train_from_scratch"] and \
    params["trainer"]["mode"] == "train":
    # deleting the previous checkpoints:
    ckp_files = glob.glob(os.path.join(params["trainer"]["model_dir"], '*'))
    for f in ckp_files:
        os.remove(f)

#Initialize the environment
env = gym.make(
    params["env"]["name"],
    params=params,
)
test_env = gym.make(params["env"]["name"], params=params)