danijar

import argparse
import collections
import re

TOKEN_REGEX = re.compile(r'[A-Za-z]+')
BLACKLIST = set([
    'pdf', 'and', 'the', 'proceedings', 'conference', 'ieee', 'for',
    'about', 'details', 'data', 'with', 'arxiv', 'preprint', 'advances'])

def tokenize(line):

danijar

import random
import time
import threading
import matplotlib.pyplot as plt


class Plot:

    def __init__(self):
        plt.ion()

danijar

Char-RNN Sample

Parameters

dataset = ArxivAbstracts(
    categories='stat.ML cs.NE cs.LG math.OC',
    keywords='neural network deep')
max_length = 50
sampling_temperature = 0.5

danijar

algorithms:
- config:
  | frame_skip: 6
  | history: 6
  | replay_capacity: 5e4
  name: DQN
  train_steps: 200000
  type: DQN
envs:
- SimpleGather-v0

danijar

keygen
uprises
non-infected
counter-weight
pipetted
quispel
rooster-whistles
moisturisers
enfeoffing
pre-adolescents

danijar

import numpy as np


class Network:

    def __init__(self, num_inputs, num_hidden, num_output,
                 init_weight_scale=0.5):
        self.w1 = np.random.normal(
            0, init_weight_scale, (num_inputs + 1, num_hidden))
        self.w2 = np.random.normal(

danijar

python -c "import gym,time;d=10000;e=gym.make('Ant-v1');s=time.time();e.reset();[e.reset() if e.step(e.action_space.sample())[2] else 0 for _ in range(d)];print(d/(time.time()-s),'FPS')"

danijar

#!/usr/bin/python3
"""Move or delete directories via your text editor.

Installation:

- Install sh.py via `pip3 install sh`.
- Save this file into a directory in your $PATH, for example `~/bin`.
"""
import argparse
import tempfile

danijar

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt


def gaussian_blur(image, diameter):
  padding = [[0, 0]] + [[(diameter) // 2, (diameter - 1) // 2]] * 2 + [[0, 0]]
  diameter = tf.to_float(diameter)
  filter_ = tf.range(-(diameter - 1) // 2, (diameter - 1) // 2 + 1)
  filter_ = tf.exp(-0.5 * filter_ ** 2 / (diameter / 4) ** 2)  # 2 stds.

danijar

def minitaur_config():
  # General
  algorithm = ppo.PPOAlgorithm
  num_agents = 10
  eval_episodes = 30
  use_gpu = False
  # Environment
  env = 'MinitaurBulletEnv-v0'
  max_length = 1000
  steps = 1e7  # 10M