# プロジェクトファイル格納用ディレクトリ作成
mkdir -p project/mnist
cd project/mnist
# このディレクトリ配下で使用するPythonのバージョンを指定
pyenv local 3.7.3
python --version
# 仮想環境を作成
python -m venv /home/acms/pyvenv/mnist
# 作成した仮想環境に切り替え
source /home/acms/pyvenv/mnist/bin/activate
# 前の記事のあと、pipが新しくなったようなので最新版に更新
pip install --upgrade pip
# 必要なライブラリをインストール
pip install chainer matplotlib pillow

wget https://github.com/chainer/chainer/archive/master.zip
unzip master.zip "chainer-master/examples/mnist/*"
mv chainer-master/examples/mnist/* .
rm -rf chainer-master/

# 学習
./train_mnist.py
# 推論
./inference.py --snapshot result/snapshot_iter_12000

[acms@apex-01-mst mnist]$ python ./train_mnist.py
Device: @numpy
# unit: 1000
# Minibatch-size: 100
# epoch: 20

Downloading from http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz...
Downloading from http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz...
Downloading from http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz...
Downloading from http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz...
epoch       main/loss   validation/main/loss  main/accuracy  validation/main/accuracy  elapsed_time
1           0.189041    0.0948515             0.943367       0.9681                    29.3833
2           0.0746062   0.0777412             0.9765         0.9755                    58.621
3           0.0469042   0.0741864             0.98535        0.9763                    89.0082
4           0.0361049   0.0718406             0.988283       0.9799                    119.898
5           0.028149    0.0790704             0.990717       0.9783                    150.946
6           0.0248317   0.0863064             0.991833       0.9776                    182.05
7           0.0206338   0.059041              0.993683       0.9848                    213.408
8           0.0182936   0.0980842             0.994          0.9778                    245.381
9           0.0137631   0.0988245             0.995817       0.9802                    277.376
10          0.0163252   0.0807811             0.99515        0.9824                    309.997
11          0.0144509   0.0907361             0.995          0.9797                    343.247
12          0.0133291   0.0825704             0.99595        0.9814                    376.936
13          0.0112155   0.0976595             0.99615        0.9803                    411.351
14          0.0128445   0.12487               0.9961         0.9776                    445.839
15          0.0108713   0.11923               0.996533       0.9782                    481.14
16          0.0097034   0.109636              0.9974         0.9796                    517.127
17          0.010692    0.102159              0.99655        0.98                      553.78
18          0.0102683   0.116668              0.996933       0.9807                    591.116
19          0.0123222   0.109196              0.996383       0.982                     628.863
20          0.00460523  0.0942767             0.998733       0.9843                    667.131
[acms@apex-01-mst mnist]$ ls
README.md     result                      train_mnist_data_parallel.py
inference.py  train_mnist.py              train_mnist_model_parallel.py
master.zip    train_mnist_custom_loop.py
[acms@apex-01-mst mnist]$ ls result/
accuracy.png  cg.dot  log  loss.png  snapshot_iter_12000
[acms@apex-01-mst mnist]$ ./inference.py --snapshot result/snapshot_iter_12000
Device: @numpy
# unit: 1000

Prediction: 7
Answer: 7
[acms@apex-01-mst mnist]$ 

cp inference.py predict_mnist.py
vi predict_mnist.py
vi predict_mnist.sh
chmod +x predict_mnist.sh

#!/usr/bin/env python
import argparse

import chainer
from train_mnist import MLP
from train_mnist_model_parallel import ParallelMLP
import numpy as np
from PIL import Image, ImageOps

def loadPredImg(path):
    # read image file and resize 28*28 grayscale
    img = Image.open(path)
    img = img.resize((28,28))
    img = img.convert('RGB')
    img = ImageOps.invert(img)
    img = img.convert('L')
    img = np.asarray(img).reshape( 1, 28, 28)
    # flatten
    img = img.reshape(784)
    # normalize
    img = img.astype(np.float32) / 255.0
    # label
    lbl = np.array(-1)

    return img, lbl

def main():
    parser = argparse.ArgumentParser(description='Chainer example: MNIST')
    parser.add_argument('--device', '-d', type=str, default='-1',
                        help='Device specifier. Either ChainerX device '
                        'specifier or an integer. If non-negative integer, '
                        'CuPy arrays with specified device id are used. If '
                        'negative integer, NumPy arrays are used')
    parser.add_argument('--snapshot', '-s',
                        default='result/snapshot_iter_12000',
                        help='The path to a saved snapshot (NPZ)')
    parser.add_argument('--unit', '-u', type=int, default=1000,
                        help='Number of units')
    parser.add_argument('--image', '-i', type=str, default='',
                       help='Image for prediction.')
    group = parser.add_argument_group('deprecated arguments')
    group.add_argument('--gpu', '-g', dest='device',
                       type=int, nargs='?', const=0,
                       help='GPU ID (negative value indicates CPU)')
    args = parser.parse_args()

    device = chainer.get_device(args.device)

    print('Device: {}'.format(device))
    print('# unit: {}'.format(args.unit))
    print('image file: ', args.image)
    print('')

    device.use()

    # Create a same model object as what you used for training
    if 'result_model_parallel' in args.snapshot:
        model = ParallelMLP(args.unit, 10, args.gpu, args.gpu)
    else:
        model = MLP(args.unit, 10)

    # Load saved parameters from a NPZ file of the Trainer object
    try:
        chainer.serializers.load_npz(
            args.snapshot, model, path='updater/model:main/predictor/')
    except Exception:
        chainer.serializers.load_npz(
            args.snapshot, model, path='predictor/')

    model.to_device(device)

    # Prepare data
    if len(args.image) > 0:
        x, answer = loadPredImg(args.image)
    else:
        train, test = chainer.datasets.get_mnist()
        x, answer = test[0]

    x = device.send(x)
    with chainer.using_config('train', False):
        prediction = model(x[None, ...])[0].array.argmax()

    print('Prediction:', prediction)
    print('Answer:', answer)


if __name__ == '__main__':
    main()

#!/bin/sh
# predict mnist
# $1 : image file name
# $2 : result file name
# $3 : model file name
# $4 : python script file name

source /home/acms/pyvenv/mnist/bin/activate
predict_value=`python $4 --snapshot $3 --image $1  |grep 'Prediction:' | cut -c 13`
deactivate

cp $1 newfile
echo section=data1 >> $2
echo predict_value=$predict_value >> $2
echo output_file=newfile >> $2
echo section=end >> $2
exit 0

mkdir predicted_images
mkdir predicted_images/0
mkdir predicted_images/1
mkdir predicted_images/2
mkdir predicted_images/3
mkdir predicted_images/4
mkdir predicted_images/5
mkdir predicted_images/6
mkdir predicted_images/7
mkdir predicted_images/8
mkdir predicted_images/9

find predicted_images/

[acms@apex-01-mst mnist]$ find predicted_images/
predicted_images/
predicted_images/0
predicted_images/1
predicted_images/2
predicted_images/2/20190715133931379000
predicted_images/3
predicted_images/4
predicted_images/5
predicted_images/6
predicted_images/7
predicted_images/8
predicted_images/9
[acms@apex-01-mst mnist]$