Inference works with Chainer 2.0 MNIST sample

Introduction

I am studying deep learning by running Chainer's example code.

At the time of writing this article (June 2017), the latest version of Chainer is 2.0, but it may not be compatible with 1.x and older versions of the code may not work. Reference: Differences between chainer versions (as of January 19, 2016)

This article is a Chainer 2.0 MNIST sample and is an implementation note for working inference.

For the implementation, I referred to this article. Chainer: Tutorial for Beginners Vol.1

environment

Chainer 2.0 python 2.7.10 Run on CPU

code

https://github.com/abechi/chainer_mnist_predict

Chainer 2.0 MNIST sample (original) https://github.com/chainer/chainer/tree/v2.0.0/examples/mnist

1. Added process (1 line) to save trained model to train_mnist.py

train_mnist.py


    # Run the training
    trainer.run()

    chainer.serializers.save_npz('my_mnist.model', model) # Added

2. Run train_mnist.py to start learning

$ python train_mnist.py --epoch 3
GPU: -1
# unit: 1000
# Minibatch-size: 100
# epoch: 3

epoch       main/loss   validation/main/loss  main/accuracy  validation/main/accuracy  elapsed_time
1           0.191836    0.0885223             0.942233       0.9718                    26.099        
2           0.0726428   0.0825069             0.9768         0.974                     53.4849       
3           0.0466335   0.0751425             0.984983       0.9747                    81.2683       
$ ls
my_mnist.model  result/         train_mnist.py*

3. Load and infer the saved trained model

predict_mnist.py


#!/usr/bin/env python

from __future__ import print_function 

try:
    import matplotlib
    matplotlib.use('Agg')
except ImportError:
    pass

import argparse

import chainer
import chainer.functions as F
import chainer.links as L
from chainer import training
from chainer.training import extensions


# Network definition
class MLP(chainer.Chain):

    def __init__(self, n_units, n_out):
        super(MLP, self).__init__()
        with self.init_scope():
            # the size of the inputs to each layer will be inferred
            self.l1 = L.Linear(None, n_units)  # n_in -> n_units
            self.l2 = L.Linear(None, n_units)  # n_units -> n_units
            self.l3 = L.Linear(None, n_out)  # n_units -> n_out

    def __call__(self, x):
        h1 = F.relu(self.l1(x))
        h2 = F.relu(self.l2(h1))
        return self.l3(h2)


def main():
    parser = argparse.ArgumentParser(description='Chainer example: MNIST')
    parser.add_argument('--unit', '-u', type=int, default=1000,
                        help='Number of units')
    args = parser.parse_args()

    print('# unit: {}'.format(args.unit))
    print('')

    # Set up a neural network
    model = L.Classifier(MLP(args.unit, 10))

    # Load the MNIST dataset
    train, test = chainer.datasets.get_mnist()

    chainer.serializers.load_npz('my_mnist.model', model)

    x, t = test[0]
    print('label:', t)

    x = x[None, ...]
    y = model.predictor(x)
    y = y.data

    print('predicted_label:', y.argmax(axis=1)[0])

if __name__ == '__main__':
    main()

predict_mnist.py reads my_mnist.model to infer labels for test data.

$ python predict_mnist.py 
# unit: 1000

label: 7
predicted_label: 7

I got the same label as the correct label.

Precautions when creating model objects

train_mnist.py


    # iteration, which will be used by the PrintReport extension below.
    model = L.Classifier(MLP(args.unit, 10))

In train_mnist.py, I made a model using L.Classifier. You need to use L.Classifier as well when creating a model object during inference.

If you create an object for your model without going through L.Classifier, you will get an error when you load the model.

predict_mnist.py


    # Set up a neural network
    model = MLP(args.unit, 10)

error KeyError: 'l2/b is not a file in the archive'

Reference Save and load Chainer model

Recommended Posts

Inference works with Chainer 2.0 MNIST sample
What to do if the Chainer (Windows) sample mnist terminates with WinError 183.
Seq2Seq (1) with chainer
MNIST (DCNN) with skflow
Use tensorboard with Chainer
Explaining mnist after chainer 1.11.0
Sample data created with python
Read the keras mnist sample
Try implementing RBM with chainer.
Learn elliptical orbits with Chainer
Seq2Seq (3) ~ CopyNet Edition ~ with chainer
Use chainer with Jetson TK1
Neural network starting with Chainer
Implemented Conditional GAN with chainer
[Python] Bayesian inference with Pyro
Implemented SmoothGrad with Chainer v2
Deep Embedded Clustering with Chainer 2.0
A little stuck with chainer
Think about dropouts with MNIST
Try TensorFlow MNIST with RNN
Works with Python and R