Test h5 model on Test Dataset-CodePudding

Using transfer learning for binary classification. I train the model using my dataset which has been split into three folders - train, test, val. Each one of these further contains the individual folders for each class.

r = model.fit_generator(
  training_set,
  validation_data = val_set, 
  epochs=5,
  steps_per_epoch=len(training_set),
  validation_steps=len(test_set)
)

After training I get the model saved as an h5 file.

import tensorflow as tf

from keras.models import load_model

model.save('vgg16_new_model.h5')

How do I use this to test the model on the test set?

CodePudding user response：

There is an equivalent to fit_generator called evaluate_generator, which you can use when you want to pass a test dataset to your trained model. However, both options are deprecated in the latest Tensorflow version, so just use model.fit and model.evaluate. Here is a simple example:

import tensorflow as tf

flowers = tf.keras.utils.get_file(
    'flower_photos',
    'https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz',
    untar=True)

img_gen = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255, rotation_range=20)

model = tf.keras.applications.vgg16.VGG16(include_top=False, input_shape=(256, 256, 3))
x = tf.keras.layers.Flatten()(model.layers[-1].output)
x = tf.keras.layers.Dense(1024, activation='relu')(x)
output = tf.keras.layers.Dense(5)(x)
model = tf.keras.Model(inputs=model.inputs, outputs=output)
model.summary()

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

epochs=1
model.fit(img_gen.flow_from_directory(flowers, batch_size=32, class_mode='sparse'),  epochs=epochs)
model.save('vgg16_new_model.h5')

##############################################################
new_model = tf.keras.models.load_model('vgg16_new_model.h5')
results = new_model.evaluate(img_gen.flow_from_directory(flowers, batch_size=32, class_mode='sparse'))
tf.print('Accuracy: ', results[1]*100)

Found 3670 images belonging to 5 classes.
115/115 [==============================] - 73s 629ms/step - loss: 1.6048 - accuracy: 0.2447
Accuracy:  24.468664824962616

Note that I am using the same subset for training and evaluation, but you would pass your test set to model.evaluate.