ubam config

windows 7 config

install git
install anaconda3
open command line (windows+r, type cmd, then enter)

mode 190,50
git clone https://github.com/wangbx66/ubam
pip install theano
pip install lasagne
pip install h5py
pip install pulp
pip install gym

close the command line and open a new command line, then test with the following and ensure no error warnings

python
import scipy
import theano

new a file .theanorc to c:\users\szheng, and write

#!sh
[global]
device=cpu
floatX=float32

[nvcc]
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin
flags=-LC:\Users\szheng\AppData\Local\Continuum\Anaconda3

To Start With

First download datasets from Game Trace Archive, and supplementary datasets from kaggle. Remove all headers and non-ascii characters (keep escaped unicodes)

##Sam

UBAM DL

Step1:

Agent.py % Parse and clean WoWH dataset % Step1: Trajs % Step2: Compress Trajs into HDF5

• reward() % calculate satisfaction, output data/trajs

• hdf_dump() % experience reply data

Architecure.py

• Buid and train Q model

Example:

python architecture.py 0 3000 300 0.0025 sam-sep-30

0 (reward index 0 means all 5f, 1 means f1 (advancement))

3000 ( batch x 24(frame/reply))

300 (loop length/horzion)

0.0025 (learning rate)

sam-sep-30 (file name for Q network)

Output file: Q Network architecture.log (for training evaluation)

Recover.py

% Load Q network % Formulate iRL cons_gen function % q_val_eval Q(st,a) % pulpsol (Python Linear Programming Solver)