pythonの構文をjuliaで書き換えたい。

Question

前提・実現したいこと

pythonでの構文をjuliaで書き換えたい。

発生している問題・エラーメッセージ

自分なりにjuliaの言語で書き換えても実行できない。

該当のソースコード

独立版Keras用コード

from future import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K

batch_size = 128
num_classes = 10
epochs = 12

input image dimensions

img_rows, img_cols = 28, 28

the data, split between train and test sets

(x_train, y_train), (x_test, y_test) = mnist.load_data()

if K.image_data_format() == 'channels_first':
x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
input_shape = (1, img_rows, img_cols)
else:
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

convert class vectors to binary class matrices

y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
activation='relu',
input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
metrics=['accuracy'])

model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[0+1]) #Blogの都合で意図的に[0+1]と表記

試したこと

using PyCall
const keras
const mnist=keras[:datasets]
const Sequential=keras[:models]
const Dense,Dropout,Flatten=keras[:layers]
const Conv2D,MaxPooling2D=keras[:layers]
const backend as K=keras
上の部分だけをこのように変更してみました。

補足情報（FW/ツールのバージョンなど）

ここにより詳細な情報を記載してください。

Comments

[y_waiwai]

このままではコードが読めないので、質問を編集し、<code>ボタンを押し、出てくる’’’の枠の中にコードを貼り付けてください

で、提示のコードではどういう不具合があるんでしょうか。
また、それをどういうふうにしたいのか、という説明も提示しましょう

Answer 1

最近はPyCallよりもPythonCallの方が推奨されているので，そちらの方での実装例です。

PythonCallの場合は，Pythonのオブジェクトを自動変換しないので，意図的に変換しないといけない部分があります。
例えば if string(K.image_data_format()) == "channels_first" のところの
K.image_data_format()はPythonの文字列型なので，うっかり
if K.image_data_format() == "channels_first"で，'channels_first'が返ってきても，違うので必ずfalse
となります。

また，Juliaでは[データ, ...]の表現は配列型， Pythonではリスト型なので，リスト型の場合明確にpylist(...)としないといけないようです。

using PythonCall

tf = pyimport("tensorflow")
np = pyimport("numpy")
keras = tf.keras
mnist = keras.datasets.mnist
Sequential = keras.models.Sequential
Dense, Dropout = keras.layers.Dense, keras.layers.Dropout
Flatten, Conv2D = keras.layers.Flatten, keras.layers.Conv2D
MaxPooling2D = keras.layers.MaxPooling2D
K = keras.backend

batch_size = 128
num_classes = 10
epochs = 12

# input image dimensions
img_rows, img_cols = 28, 28

# the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

if string(K.image_data_format()) == "channels_first"
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)    
    input_shape = (1, img_rows, img_cols);
else
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1);
end

x_train /= 255.0
x_test  /= 255.0
println("x_train shape: $(np.shape(x_train))")
println("$(pylen(x_train)) train samples")
println("$(pylen(x_test)) test samples")

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential();
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation="relu",
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=pylist(["accuracy"]))

model.fit(x=x_train, y=y_train,
          batch_size=batch_size,
          epochs=epochs, 
          validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test)
println("Test loss: $(score[0])")
println("Test accuracy: $(score[1])")

Answer 2

最新のPyCallだとよりpythonとの区別がつかない書き方ができるようですね。
次のような感じでどうでしょうか?

using PyCall

const tensorflow = pyimport("tensorflow")
const keras = tensorflow.keras
const mnist = keras.datasets.mnist
const Sequential = keras.models.Sequential
const Dense, Dropout = keras.layers.Dense, keras.layers.Dropout
const Flatten, Conv2D = keras.layers.Flatten, keras.layers.Conv2D
const MaxPooling2D = keras.layers.MaxPooling2D
const K = keras.backend

batch_size = 128
num_classes = 10
epochs = 12

# input image dimensions
img_rows, img_cols = 28, 28

# the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

if K.image_data_format() == "channels_first"
    x_train = reshape(x_train, (size(x_train)[1], 1, img_rows, img_cols));
    x_test = reshape(x_test, (size(x_test)[1], 1, img_rows, img_cols));
    input_shape = (1, img_rows, img_cols);
else
    x_train = reshape(x_train, (size(x_train)[1], img_rows, img_cols, 1));
    x_test = reshape(x_test, (size(x_test)[1], img_rows, img_cols, 1));
    input_shape = (img_rows, img_cols, 1);
end

x_train = Float32.(x_train ./ 255.0) 
x_test  = Float32.(x_test ./ 255.0) 
println("x_train shape: $(size(x_train))")
println("$(size(x_train)[1]) train samples")
println("$(size(x_test)[1]) test samples")

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential();
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation="relu",
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=["accuracy"])

model.fit(x=x_train, y=y_train,
          batch_size=batch_size,
          epochs=epochs, 
          verbose=1,
          validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
println("Test loss: $(score[1])")
println("Test accuracy: $(score[1+1])")

動作環境としては次の通りです。

• ubuntu 22.04 on x86_64
• Julia 1.8.0
• Conda(Juliaのパッケージ) 1.7.0
• PyCall 1.94.1

既存の状態だとconst tensorflow = pyimport("tensorflow")でエラーが出たため

using Pkg
Pkg.remove("Conda")
Pkg.remove("PyCall")

の後，

rm -rf $HOME/.julia/conda

でJulia環境内のCondaを全消し。その後

using Pkg
Pkg.add("Conda")
Pkg.add("PyCall")
using Conda
Conda.update()
Conda.add("tensorflow")
Pkg.build("PyCall")

のような感じで環境を整えました。