mirror of
https://github.com/vale981/ray
synced 2025-03-06 18:41:40 -05:00
60d4d5e1aa
* Remove all __future__ imports from RLlib. * Remove (object) again from tf_run_builder.py::TFRunBuilder. * Fix 2xLINT warnings. * Fix broken appo_policy import (must be appo_tf_policy) * Remove future imports from all other ray files (not just RLlib). * Remove future imports from all other ray files (not just RLlib). * Remove future import blocks that contain `unicode_literals` as well. Revert appo_tf_policy.py to appo_policy.py (belongs to another PR). * Add two empty lines before Schedule class. * Put back __future__ imports into determine_tests_to_run.py. Fails otherwise on a py2/print related error.
59 lines
1.6 KiB
Python
59 lines
1.6 KiB
Python
# Code in this file is copied and adapted from
|
|
# https://github.com/openai/evolution-strategies-starter.
|
|
|
|
import numpy as np
|
|
from ray.rllib.utils import try_import_tf
|
|
|
|
tf = try_import_tf()
|
|
|
|
|
|
def compute_ranks(x):
|
|
"""Returns ranks in [0, len(x))
|
|
|
|
Note: This is different from scipy.stats.rankdata, which returns ranks in
|
|
[1, len(x)].
|
|
"""
|
|
assert x.ndim == 1
|
|
ranks = np.empty(len(x), dtype=int)
|
|
ranks[x.argsort()] = np.arange(len(x))
|
|
return ranks
|
|
|
|
|
|
def compute_centered_ranks(x):
|
|
y = compute_ranks(x.ravel()).reshape(x.shape).astype(np.float32)
|
|
y /= (x.size - 1)
|
|
y -= 0.5
|
|
return y
|
|
|
|
|
|
def make_session(single_threaded):
|
|
if not single_threaded:
|
|
return tf.Session()
|
|
return tf.Session(
|
|
config=tf.ConfigProto(
|
|
inter_op_parallelism_threads=1, intra_op_parallelism_threads=1))
|
|
|
|
|
|
def itergroups(items, group_size):
|
|
assert group_size >= 1
|
|
group = []
|
|
for x in items:
|
|
group.append(x)
|
|
if len(group) == group_size:
|
|
yield tuple(group)
|
|
del group[:]
|
|
if group:
|
|
yield tuple(group)
|
|
|
|
|
|
def batched_weighted_sum(weights, vecs, batch_size):
|
|
total = 0
|
|
num_items_summed = 0
|
|
for batch_weights, batch_vecs in zip(
|
|
itergroups(weights, batch_size), itergroups(vecs, batch_size)):
|
|
assert len(batch_weights) == len(batch_vecs) <= batch_size
|
|
total += np.dot(
|
|
np.asarray(batch_weights, dtype=np.float32),
|
|
np.asarray(batch_vecs, dtype=np.float32))
|
|
num_items_summed += len(batch_weights)
|
|
return total, num_items_summed
|