epignatelli

"""
This gist is an extract from:
https://github.com/epignatelli/reinforcement-learning-an-introduction
"""

    def bellman_expectation(self, state, probs, discount):
        """
        Makes a one step lookahead and applies the bellman expectation equation to the state self.state_value[state]
        Args:
            state (Tuple[int, int]): the x, y indices that define the address on the value table

epignatelli

"""
This gist is an extract from:
https://github.com/epignatelli/reinforcement-learning-an-introduction
"""

import matplotlib
matplotlib.use("Agg")

import numpy as np
import matplotlib.pyplot as plt

epignatelli

"""
An implementation of the pytorch Subset that returns an instance of the original dataset with a reduced number of items.
This has two benefits:
  - It allows to stil access the attributes of the Dataset class, such as methods, or properties.
  - You can use the usual python index notation with slices to chunk the dataset, rather than creating a list of indices

"""
class Dataset(object):
  def __init__(self, iterable):
    self.items = iterable

epignatelli

import torch

def is_supported():
    CUDA_VERSION = torch._C._cuda_getCompiledVersion()
    supported = True
    for d in range(torch.cuda.device_count()):
        capability = torch.cuda.get_device_capability(d)
        major = capability[0]
        minor = capability[1]
        supported &= major > 3  # too old

epignatelli

import hashlib

funcs = [
    hashlib.blake2b,
    hashlib.blake2s,
    hashlib.md5,
    hashlib.sha1,
    hashlib.sha224,
    hashlib.sha256,
    hashlib.sha256,

epignatelli

def profiled(func):
    def decorated(*args, **kwargs):
        pr = cProfile.Profile()
        pr.enable()
        result = func(*args, **kwargs)
        pr.disable()
        pr.print_stats(sort="tottime")
        return result

epignatelli

from torch.utils.data import Subset
from sklearn.model_selection import train_test_split


class SubsetProxy(Subset):
    def __init__(self, dataset, indices):
        super().__init__(dataset, indices)

    def __getattr__(self, name):
        try:

epignatelli

import math


def conv1d_out_shape(input_size, kernel_size, stride=1, padding=0, dilation=1):
    output_size = ((input_size + (2 * padding[0]) - dilation[0] * (kernel_size[0] - 1) - 1) / stride[0]) + 1
    return math.floor(output_size)


def transpose_conv1d_out_shape(input_size, kernel_size, stride=1, padding=0, dilation=1):
    output_size = ((input_size - 1) * stride) - (2 * padding) + (dilation * (kernel_size - 1))

epignatelli

import sys, functools

def private(member):
    @functools.wraps(member)
    def wrapper(*function_args):
      myself = member.__name__
      caller = sys._getframe(1).f_code.co_name
      if (not caller in dir(function_args[0]) and not caller is myself):
         raise PermissionError(f"Cannot call private function {caller}.{myself}")
      return member(*function_args)