gngdb

# written by Harri Edwards, edited by Gavin Gray

import pickle
import tensorflow as tf
import numpy as np
import tf_util
import gym
import load_policy
import gym
import tensorflow.contrib.slim as slim

gngdb

def value_iteration(mdp, gamma, nIt):
    """
    Inputs:
        mdp: MDP
        gamma: discount factor
        nIt: number of iterations, corresponding to n above
    Outputs:
        (value_functions, policies)
        
    len(value_functions) == nIt+1 and len(policies) == n

gngdb

gngdb

gngdb

gngdb

import math

def sgdr(period, batch_idx):
    # returns normalised anytime sgdr schedule given period and batch_idx
    # best performing settings reported in paper are T_0 = 10, T_mult=2
    # so always use T_mult=2
    batch_idx = float(batch_idx)
    restart_period = period
    while batch_idx/restart_period > 1.:
        batch_idx = batch_idx - restart_period

gngdb

import sys
try:
    import pyPdf
except ImportError:
    import PyPDF2 as pyPdf

def addPage(opdf, ipdf, pg):
    if pg>=len(pages):
        opdf.addBlankPage(w, h)
    else:

gngdb

import tensorflow as tf
import autograd.numpy as np
from autograd import grad
from tensorflow.python.framework import function

rng = np.random.RandomState(42)
x_np = rng.randn(4,4).astype(np.float32)
with tf.device('/cpu:0'):
    x = tf.Variable(x_np)

gngdb

import tensorflow as tf
from tensorflow.python.framework import function

@function.Defun()
def my_op_grad(op, grad): ### instead of my_op_grad(x)
    return tf.sigmoid(op)

@function.Defun(grad_func=my_op_grad)
def my_op(a):
    return tf.identity(a)

gngdb