'''Script to go with AppNexus blog post.

Taken from and altered from:

  classifier_from_little_data_script_1.py
  https://gist.github.com/fchollet/0830affa1f7f19fd47b06d4cf89ed44d

which appears at

  https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html
  by Francois Chollet

It uses data that can be downloaded at:

  https://www.kaggle.com/c/dogs-vs-cats/data

The rest of this file was written by

  Ryan Woodard | AppNexus | Data Science | 2017
'''

#
# Original code from Francois Chollet, Keras
#
import keras
from keras import backend as K
from keras.layers import Input, Dense, Conv2D, MaxPooling2D, Flatten, Dropout
from keras.models import Model

# dimensions of our images.
img_width, img_height = 150, 150

train_data_dir = 'data/train'
validation_data_dir = 'data/validation'
nb_train_samples = 2000
nb_validation_samples = 800
epochs = 5  #50
batch_size = 16


#
# left branch of model (convnet)
#
if K.image_data_format() == 'channels_first':
    input_shape = (3, img_width, img_height)
else:
    input_shape = (img_width, img_height, 3)

# This returns a tensor
linput = Input(shape=input_shape)

x = Conv2D(32, (3, 3), padding='same', activation='relu')(linput)
x = MaxPooling2D((2, 2))(x)

x = Conv2D(32, (3, 3), padding='same', activation='relu')(x)
x = MaxPooling2D((2, 2))(x)

x = Conv2D(64, (3, 3), padding='same', activation='relu')(x)
x = MaxPooling2D((2, 2))(x)

loutput = Flatten()(x)


#
# right branch of model (simple feature data, design matrix)
#
nfeatures = 74  # From akmtdfgen.py test_generator()
rinput = Input(shape=(nfeatures,), name='rinput')


#
# Make the merged model.
#
x = keras.layers.concatenate([loutput, rinput])
x = Dense(64, activation='relu')(x)
x = Dense(64, activation='relu')(x)
x = Dropout(0.5)(x)

# And finally we add the main logistic regression layer
main_output = Dense(1, activation='sigmoid', name='main_output')(x)

model = Model(inputs=[linput, rinput], outputs=main_output)

model.compile(loss='binary_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])


from akmtdfgen import get_demo_data
from akmtdfgen import generator_from_df

df_train, df_valid = get_demo_data()

assert {2} == set([df_train.target.nunique(),
                   df_train.label.nunique(),
                   df_valid.target.nunique(),
                   df_valid.label.nunique()])

ntrain, nvalid = df_train.shape[0], df_valid.shape[0]

print("""
Training set: %d images, 2 classes.
Validation set: %d images, 2 classes.
""" % (ntrain, nvalid))


# lmodel.fit(data, labels)  # starts training

target_size = (img_width, img_height)
train_generator = generator_from_df(df_train, batch_size, target_size, features="mm_features_train_bc")
validation_generator = generator_from_df(df_valid, batch_size, target_size, features="mm_features_valid_bc")

nbatches_train, mod = divmod(ntrain, batch_size)
nbatches_valid, mod = divmod(nvalid, batch_size)

nworkers = 10

# Latest Keras 2.0 API:
# fit_generator(self, generator, steps_per_epoch, epochs=1, verbose=1,
#               callbacks=None, validation_data=None, validation_steps=None,
#               class_weight=None, max_queue_size=10, workers=1,
#               use_multiprocessing=False, initial_epoch=0)


model.fit_generator(
    train_generator,
    steps_per_epoch=nbatches_train,
    epochs=epochs,
    verbose=2,
    validation_data=validation_generator,
    validation_steps=nbatches_valid,
    workers=nworkers)


#lmodel.save_weights('mm_mt_df_gen.h5')