動作環境

GeForce GTX 1070 (8GB)
ASRock Z170M Pro4S [Intel Z170chipset]
Ubuntu 14.04 LTS desktop amd64
TensorFlow v0.11
cuDNN v5.1 for Linux
CUDA v8.0
Python 2.7.6
IPython 5.1.0 -- An enhanced Interactive Python.
gcc (Ubuntu 4.8.4-2ubuntu1~14.04.3) 4.8.4
GNU bash, version 4.3.8(1)-release (x86_64-pc-linux-gnu)

概要

http://qiita.com/7of9/items/09262a2ab01d037d169b
にて学習したネットワークをもとに、入力データを与えてpredictionに相当するものを出力する。

後々の画像化を考慮して、入力データのファイル（input.csv)と同じ形式で出力するようにした。

code v0.1

reproduce_170429.py

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import math
import sys

'''
v0.1 Apr. 29, 2017
	- output by replacing the learned [Exr]
    - branched from [check_resultmap_170329.ipynb]
'''

# to ON/OFF debug print at one place
def output_debugPrint(str):
    #print(str)
    pass  # no operation

def calc_sigmoid(x):
    return 1.0 / (1.0 + math.exp(-x))


def calc_conv(src, weight, bias, applyActFnc):
    wgt = weight.shape
#   print wgt # debug
    #conv = list(range(bias.size))
    conv = [0.0] * bias.size

    # weight
    for idx1 in range(wgt[0]):
        for idx2 in range(wgt[1]):
            conv[idx2] = conv[idx2] + src[idx1] * weight[idx1, idx2]
    # bias
    for idx2 in range(wgt[1]):
        conv[idx2] = conv[idx2] + bias[idx2]
    # activation function
    if applyActFnc:
        for idx2 in range(wgt[1]):
            conv[idx2] = calc_sigmoid(conv[idx2])
    return conv  # return list

model_var = np.load('model_variables_170429.npy')
learn_data = np.loadtxt('input.csv', delimiter=',')

prd_data = np.array([])

for idx, items in enumerate(learn_data):
    if len(items) < 1:  # avoid empty line
        break
    #if idx >= 5:  # for debug
    #    break
    print(idx)
    # input layer (3 node)
    inlist = (items[0], items[1], items[2])
    # hidden layer 1
    outdata = calc_conv(inlist, model_var[0], model_var[1], applyActFnc=True)
    # hidden layer 2
    outdata = calc_conv(outdata, model_var[2], model_var[3], applyActFnc=True)
    # hidden layer 3
    outdata = calc_conv(outdata, model_var[4], model_var[5], applyActFnc=True)
    # output layer
    outdata = calc_conv(outdata, model_var[6], model_var[7], applyActFnc=False)
    
    wrk = [ 
    	   items[0], items[1], items[2],  # x,y,z
    	   outdata[0],  items[4],  # Exr, Exi
    	   items[5], items[6],  # Eyr, Eyi
    	   items[7], items[8]   # Ezr, Ezi
    	   ]
    #print(wrk)
    prd_data = np.append(prd_data, wrk)

print('idx=',idx)

NUM_ITEMS = 9
ncols = len(prd_data) / NUM_ITEMS
prd_data = np.array(prd_data).reshape(ncols, NUM_ITEMS)

np.savetxt("input_Exr_replaced_170429.csv", prd_data, delimiter=","
	       , fmt="%.9f")

結果

$ head input_Exr_replaced_170429.csv 
-0.209439510,-1.466076572,-5.235987756,0.068316654,0.113688322,0.519136168,0.356408494,-0.109750412,0.005652638
0.209439510,-1.466076572,-5.235987756,0.088643790,0.075268157,0.806765822,0.313605899,-0.087339272,0.002273401
-1.047197551,-1.047197551,-5.235987756,0.025312938,0.105477041,-0.151919209,0.284581139,-0.040284662,0.014306838
-0.628318531,-1.047197551,-5.235987756,0.040992646,0.090998304,0.132621814,0.389916050,-0.049129976,0.030650874
-0.209439510,-1.047197551,-5.235987756,0.050157650,0.044499152,0.422640054,0.416238663,-0.054321067,0.030099672
0.209439510,-1.047197551,-5.235987756,0.058198942,0.081592287,0.696098762,0.228374544,-0.029275715,0.000805544
0.628318531,-1.047197551,-5.235987756,0.065595694,-0.026843276,0.797857025,0.040918210,-0.023928849,-0.034062430
1.047197551,-1.047197551,-5.235987756,0.065374163,-0.055698835,0.728159227,-0.100947639,-0.034184970,-0.041232617
-1.047197551,-0.628318531,-5.235987756,0.016307764,0.083149606,-0.275568958,0.256085316,-0.040395470,0.006757636
-0.628318531,-0.628318531,-5.235987756,0.027492900,0.041832652,0.087782092,0.394477943,-0.037243314,0.020279789

遅い

numpyをうまく使えていないので、すごく遅い。

9327行の処理で３分かかる。

real	2m57.543s
user	2m56.590s
sys	0m0.485s

numpyの扱いをもっと学ばないといけない。