Load model failes in Keras for this model

SampleClassifier.py

import datetime
import os
from time import time

import keras
import numpy as np
from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
from keras.preprocessing.image import ImageDataGenerator
from keras.layers import Activation, BatchNormalization, Convolution2D, Dense, Dropout, Flatten, MaxPooling2D
from keras.models import Sequential
from keras.optimizers import SGD
from keras.regularizers import l2


best_model_path = "test.h5"


def add_convolution(classifier, filters, kernel_size, 0.0001, strides=(1, 1), input_shape=None):
    if input_shape is None:
        classifier.add(Convolution2D(filters, kernel_size, strides=strides, padding='same',
                                     kernel_regularizer=l2(0.0001)))
    else:
        classifier.add(
                Convolution2D(filters, kernel_size, padding='same', kernel_regularizer=l2(0.0001),
                              input_shape=input_shape))
    classifier.add(BatchNormalization())
    classifier.add(Activation('relu'))
	


classifier = Sequential()

add_convolution(classifier, 16, 3, 0.0001, input_shape=data_shape)
add_convolution(classifier, 16, 3, 0.0001)
classifier.add(MaxPooling2D())

add_convolution(classifier, 32, 3, 0.0001)
add_convolution(classifier, 32, 3, 0.0001)
classifier.add(MaxPooling2D())

add_convolution(classifier, 64, 3, 0.0001)
add_convolution(classifier, 64, 3, 0.0001)
add_convolution(classifier, 64, 3, 0.0001)
classifier.add(MaxPooling2D())

add_convolution(classifier, 128, 3, 0.0001)
add_convolution(classifier, 128, 3, 0.0001)
add_convolution(classifier, 128, 3, 0.0001)
classifier.add(MaxPooling2D())

add_convolution(classifier, 192, 3, 0.0001)
add_convolution(classifier, 192, 3, 0.0001)
add_convolution(classifier, 192, 3, 0.0001)
add_convolution(classifier, 192, 3, 0.0001)
classifier.add(MaxPooling2D())

classifier.add(Flatten())  # Flatten
classifier.add(Dropout(0.5))
classifier.add(Dense(units=2, kernel_regularizer=l2(0.0001)))
classifier.add(Activation('softmax', name="output_node"))

stochastic_gradient_descent = SGD(lr=0.0001, momentum=0.9, nesterov=True)
classifier.compile(stochastic_gradient_descent, loss="categorical_crossentropy", metrics=["accuracy"])

model_checkpoint = ModelCheckpoint(best_model_path, monitor="val_acc", save_best_only=True, verbose=1)

history = model.fit_generator(
        generator=training_data_generator,
        steps_per_epoch=training_steps_per_epoch,
        epochs=training_configuration.number_of_epochs,
        callbacks=[model_checkpoint],
        validation_data=validation_data_generator,
        validation_steps=validation_steps_per_epoch
)

# For some models, loading the model directly does not work, but loading the weights does
# (see https://github.com/fchollet/keras/issues/4044#issuecomment-254921595)
best_model = keras.models.load_model(best_model_path)