У Кераса странные проблемы с обученными моделями зоопарка Densenet / Resnet на MNIST

Question

Я невероятно озадачен и надеюсь, что есть какое-то простое решение. Очевидный лайнер, который я забыл. Результаты обучения и результаты оценки / прогнозирования не совпадают для моделей зоопарка библиотеки Keras. Выполнено на Keras 2.3.0.

TLDR: результаты обучения для модельного зоопарка Densenet:

model.fit(x_train2[:128], y_train[:128],
      batch_size=batch_size,
      epochs=40,
      verbose=1,)
Epoch 40/40
128/128 [==============================] - 0s 648us/step - loss:   1.1384e-04 - accuracy: 1.0000

Результаты прогноза:

model.evaluate(x_train2[:128], y_train[:128], verbose=0)
[2.4420450925827026, 0.1015625]

Тот же эксперимент, выполненный с простымconvnet:

model.fit(x_train[:128], y_train[:128],
      batch_size=batch_size,
      epochs=40,
      verbose=1,)
Epoch 40/40
128/128 [==============================] - 0s 38us/step - loss: 0.0245 - accuracy: 1.0000

Результаты прогноза:

model.evaluate(x_train[:128], y_train[:128], verbose=0)
[0.0009860699210548773, 1.0]

Далее приведен прогнозируемый результат модели Denset:

out = model.predict(x_train2[:128])
np.argmax(out,axis=1)
array([7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7])

Есть ли какой-то переключательнужно сказать модели на основе библиотеки Keras, что она находится в режиме логического вывода?

Далее следует полный воспроизводимый код:

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

from keras.applications.densenet import DenseNet121
import numpy as np

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)


### Simple model
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.Adam(),
              metrics=['accuracy'])

model.fit(x_train[:128], y_train[:128],
          batch_size=batch_size,
          epochs=40,
          verbose=1,)
          #validation_data=(x_test, y_test))

model.evaluate(x_train[:128], y_train[:128], verbose=0)


model = DenseNet121(include_top=True, 
         weights=None, 
         input_tensor=None, 
         input_shape=(32,32,1), 
         pooling=None, 
         classes=10)

model.compile(loss='categorical_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

x_train2 = np.zeros((128,32,32,1))
x_train2[:128,:28,:28] = x_train[:128]

model.fit(x_train2[:128], y_train[:128],
          batch_size=batch_size,
          epochs=40,
          verbose=1,)
          #validation_data=(x_test, y_test))

model.evaluate(x_train2[:128], y_train[:128], verbose=0)