File size: 12,557 Bytes
5960497 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 |
import math
import sys
if not sys.warnoptions:
import warnings
warnings.simplefilter("ignore")
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category=RuntimeWarning)
warnings.simplefilter("ignore")
import tensorflow as tf
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
tf.get_logger().setLevel('INFO')
tf.autograph.set_verbosity(0)
import warnings
warnings.filterwarnings("ignore", message=r"Passing", category=FutureWarning)
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.simplefilter(action='ignore', category=Warning)
from baselines.ppo2 import sppo2
from baselines.common.models import build_skill_impala_cnn
from baselines.common.mpi_util import setup_mpi_gpus
from procgen import ProcgenEnv
from baselines.common.vec_env import (
VecExtractDictObs,
VecMonitor,
VecNormalize
)
from baselines import logger
from mpi4py import MPI
import argparse
import matplotlib.colors as c
import gym
from PIL import Image
import matplotlib
import matplotlib.pyplot as plt
from baselines.common import set_global_seeds
from train_procgen.utils import *
from copy import deepcopy
from matplotlib.offsetbox import OffsetImage, AnnotationBbox
parser = argparse.ArgumentParser(description='Process procgen enjoying arguments.')
parser.add_argument('--env_name', type=str, default='coinrun')
parser.add_argument('--num_envs', type=int, default=64)
parser.add_argument('--distribution_mode', type=str, default='easy',
choices=["easy", "hard", "exploration", "memory", "extreme"])
parser.add_argument('--num_levels', type=int, default=0)
parser.add_argument('--start_level', type=int, default=0)
parser.add_argument('--test_worker_interval', type=int, default=0)
parser.add_argument('--timesteps_per_proc', type=int, default=25_000_000)
parser.add_argument('--rand_seed', type=int, default=2021)
parser.add_argument('--num_embeddings', type=int, default=8)
parser.add_argument('--beta', type=float, default=0.0)
parser.add_argument('--alpha1', type=float, default=20)
parser.add_argument('--alpha2', type=float, default=20)
parser.add_argument('--total_states', type=int, default=5000, help='choose how many states you want to have')
args = parser.parse_args()
def isEmpty(path):
if os.path.exists(path) and not os.path.isfile(path):
# Checking if the directory is empty or not
if not os.listdir(path):
return True
else:
return False
else:
return True
def load_model(env_name, num_envs, distribution_mode, num_levels, start_level,
num_embeddings, beta, alpha1, alpha2,
log_dir='./checkpoints', comm=None, rand_seed=None):
ent_coef = .01
nsteps = 256
nminibatches = 8
mpi_rank_weight = 1
num_levels = num_levels
logger.info("creating environment")
venv = ProcgenEnv(num_envs=num_envs, env_name=env_name, num_levels=num_levels, start_level=start_level + 200000,
distribution_mode=distribution_mode, rand_seed=rand_seed)
venv = VecExtractDictObs(venv, "rgb")
venv = VecMonitor(
venv=venv, filename=None, keep_buf=100,
)
venv = VecNormalize(venv=venv, ob=False)
logger.info("creating tf session")
setup_mpi_gpus()
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True # pylint: disable=E1101
sess = tf.compat.v1.Session(config=config)
sess.__enter__()
conv_fn = lambda x: build_skill_impala_cnn(x, depths=[16, 32, 32], emb_dim=256, num_embeddings=num_embeddings,
beta=beta, seed=rand_seed)
logger.info("loading the model")
model = sppo2.load(
env=venv,
network=conv_fn,
nsteps=nsteps,
nminibatches=nminibatches,
ent_coef=ent_coef,
mpi_rank_weight=mpi_rank_weight,
comm=comm,
vf_coef=0.5,
max_grad_norm=0.5,
seed=rand_seed,
load_path=log_dir,
alpha1=alpha1,
alpha2=alpha2,
)
return model
def main():
cluster_path = './train_procgen/figures/' + args.env_name + '_cluster_images'
# if empty, we need to collect states first
if isEmpty(cluster_path):
plt.rcParams.update({'font.size': 15})
comm = MPI.COMM_WORLD
saved_dir = './train_procgen/checkpoints/sppo-' + args.env_name + '_' + args.distribution_mode + '_' + str(
args.num_levels) + '_' + str(args.start_level) + '_' + str(
args.rand_seed)
saved_dir += '/checkpoints/' + str(args.timesteps_per_proc)
model = load_model(args.env_name,
1,
args.distribution_mode,
args.num_levels,
args.start_level,
args.num_embeddings,
comm=comm,
log_dir=saved_dir,
rand_seed=args.rand_seed,
beta=args.beta,
alpha1=args.alpha1,
alpha2=args.alpha2,
)
if not os.path.isdir(cluster_path):
if not os.path.isdir('./train_procgen/figures'):
os.mkdir('./train_procgen/figures')
os.mkdir(cluster_path)
for i in range(args.num_embeddings):
if not os.path.isdir(cluster_path + '/cluster_' + str(i)):
os.mkdir(cluster_path + '/cluster_' + str(i))
# generate the skill clusters
set_global_seeds(args.rand_seed)
env_name = "procgen:procgen-" + args.env_name + "-v0"
env = gym.make(env_name, num_levels=args.num_levels, start_level=args.start_level,
distribution_mode=args.distribution_mode, rand_seed=args.rand_seed, render_mode="human")
obs = env.reset()
counters = np.zeros(args.num_embeddings)
episode_i = 0
total_states = args.total_states
upper_frames = args.total_states * 0.2
episode_files = []
sequence_files = [[] for _ in range(args.num_embeddings)]
sequence_indices = [[] for _ in range(args.num_embeddings)]
episode_rewards = 0
episode_counters = np.zeros(args.num_embeddings)
# avoid to collect the endless loop
if args.env_name not in ['starpilot', 'fruitbot']:
episode_max_steps = 200
else:
episode_max_steps = math.inf
while True:
rgb_img = env.render(mode="rgb_array")
im = Image.fromarray(rgb_img)
a, v, pure_latent, vq_latent, pure_vq_latent, vq_embeddings, encoding_indices, sl, lat = model.skill_step(
obs)
action = np.squeeze(a)
if episode_counters[encoding_indices[0][0]] < upper_frames:
saved_dir = cluster_path + '/cluster_' + str(encoding_indices[0][0])
file_name = str(episode_i + 1) + '_' + str(sl[0][0]) + '_' + str(sl[0][1]) + '.png'
im.save(saved_dir + '/' + file_name, dpi=(5, 5))
episode_files.append(saved_dir + '/' + file_name)
if len(sequence_files[encoding_indices[0][0]]) == 0:
sequence_files[encoding_indices[0][0]].append(saved_dir + '/' + file_name)
sequence_indices[encoding_indices[0][0]].append(episode_i + 1)
else:
if (episode_i + 1 - int(sequence_indices[encoding_indices[0][0]][-1])) == 1:
sequence_files[encoding_indices[0][0]].append(saved_dir + '/' + file_name)
sequence_indices[encoding_indices[0][0]].append(episode_i + 1)
else:
if len(sequence_files[encoding_indices[0][0]]) > 20:
for f in sequence_files[encoding_indices[0][0]]:
if os.path.isfile(f):
os.remove(f)
episode_counters[encoding_indices[0][0]] -= 1
else:
continue
sequence_files[encoding_indices[0][0]] = []
sequence_indices[encoding_indices[0][0]] = []
sequence_files[encoding_indices[0][0]].append(saved_dir + '/' + file_name)
sequence_indices[encoding_indices[0][0]].append(episode_i + 1)
obs, reward, done, info = env.step(action)
episode_rewards += reward
episode_counters[encoding_indices[0][0]] += 1
episode_i += 1
if done:
if len(episode_files) > episode_max_steps:
for f in episode_files:
if os.path.isfile(f):
os.remove(f)
for i in range(args.num_embeddings):
if f in sequence_files[i]:
sequence_indices[i].pop(sequence_files[i].index(f))
sequence_files[i].remove(f)
else:
continue
episode_counters = deepcopy(counters)
counters = deepcopy(episode_counters)
episode_files = []
episode_rewards = 0
if np.sum(counters) >= total_states:
env.close()
break
skill_latents = []
cluster_indices = []
images = []
for i in range(args.num_embeddings):
files = os.listdir(cluster_path + '/cluster_' + str(i))
for file in files:
if file.endswith('png'):
_, sl_x, sl_y = file[:-4].split('_')
arr_img = plt.imread(cluster_path + '/cluster_' + str(i) + '/' + file)
images.append(arr_img)
skill_latents.append([float(sl_x), float(sl_y)])
cluster_indices.append(i)
images = np.array(images)
skill_latents = np.array(skill_latents)
cluster_indices = np.array(cluster_indices)
fig, ax = plt.subplots()
fig.set_figheight(6)
fig.set_figwidth(8)
np.random.seed(0)
color_set = [(np.random.rand(), np.random.rand(), np.random.rand()) for _ in range(args.num_embeddings)]
c_cet = []
for cs in color_set:
c_cet.append(c.to_hex(cs))
c_cet = np.array(c_cet)
figure = plt.scatter(skill_latents[:, 0], skill_latents[:, 1], s=10, c=c_cet[cluster_indices], alpha=0.5)
line, = plt.plot(skill_latents[:, 0], skill_latents[:, 1], ls="")
plt.title('The FDR space')
plt.xlabel('FDR x')
plt.ylabel('FDR y')
# create the annotations box
im = OffsetImage(images[0, :, :, :], zoom=0.2)
xybox = (100., 100.)
ab = AnnotationBbox(im, (0, 0), xybox=xybox, xycoords='data',
boxcoords="offset points", pad=0.3, arrowprops=dict(arrowstyle="->"))
# add it to the axes and make it invisible
ax.add_artist(ab)
ab.set_visible(False)
def hover(event):
# if the mouse is over the scatter points
if line.contains(event)[0]:
# find out the index within the array from the event
try:
ind, = line.contains(event)[1]["ind"]
except:
indices = line.contains(event)[1]["ind"]
ind = indices[0]
# get the figure size
w, h = fig.get_size_inches() * fig.dpi
ws = (event.x > w / 2.) * -1 + (event.x <= w / 2.)
hs = (event.y > h / 2.) * -1 + (event.y <= h / 2.)
# if event occurs in the top or right quadrant of the figure,
# change the annotation box position relative to mouse.
ab.xybox = (xybox[0] * ws, xybox[1] * hs)
# make annotation box visible
ab.set_visible(True)
# place it at the position of the hovered scatter point
ab.xy = (skill_latents[:, 0][ind], skill_latents[:, 1][ind])
# set the image corresponding to that point
im.set_data(images[ind, :, :, :])
else:
# if the mouse is not over a scatter point
ab.set_visible(False)
fig.canvas.draw_idle()
# add callback for mouse moves
fig.canvas.mpl_connect('motion_notify_event', hover)
plt.show()
if __name__ == '__main__':
main()
|