johnny5550822

""" Trains an agent with (stochastic) Policy Gradients on Pong. Uses OpenAI Gym. """
import numpy as np
import cPickle as pickle
import gym

# hyperparameters
H = 200 # number of hidden layer neurons
batch_size = 10 # every how many episodes to do a param update?
learning_rate = 1e-4
gamma = 0.99 # discount factor for reward

johnny5550822

-- ###############

-- a version with GPU support. Assuming your gpu is working properly. 

-- This is a minimal implementation (e.g., no batch processing) of a CNN on digit (0-9) classification. This code is largely obtained from luacnn, https://github.com/hpenedones/luacnn
-- The training data can be also obtained in https://github.com/hpenedones/luacnn

-- ###############

require "torch"

johnny5550822

Can you imagine a computer can generate an nba article?

The following is some nba articles fully-automatically generated by char-cnn, a recurrent-neural-network library thanks to Andrej Karpathy [link]. The library is awesome to easy, and very user-friendly. You should try it! :)

Basically, I wrote a python script [link] to extract past archives . And use that as the training set for the recurrent neural network.

The articles below are generated by a network trained with rougly about 2 millions character (which is an okay size; not big enough though). You can see that the generated article contains artificial author names, speeches, etc; similar to an nba archive (although the logic has to be improved, it is FUN.)

You can tune the parameter and train with a even bigger dataset using my script. And you will probably get better result! Have fun:)

johnny5550822

# A simple script to parse ESPN nba archive in a specified year with specified months, Code updated: 08/20/2015

import urllib2
from bs4 import BeautifulSoup # a very good python library for parsing html
import time
import re

# parameters
year = 2015
months = ['january','August']	# the service is only available in these two months

johnny5550822

"""
This is a batched LSTM forward and backward pass
"""
import numpy as np
import code

class LSTM:
  
  @staticmethod
  def init(input_size, hidden_size, fancy_forget_bias_init = 3):

johnny5550822

"""
Minimal character-level Vanilla RNN model. Written by Andrej Karpathy (@karpathy)
BSD License
"""
import numpy as np

# data I/O
data = open('input.txt', 'r').read() # should be simple plain text file
chars = list(set(data))
data_size, vocab_size = len(data), len(chars)