#  Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
#  Licensed under the Apache License, Version 2.0 (the "License");
#  you may not use this file except in compliance with the License.
#  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
#  Unless required by applicable law or agreed to in writing, software
#  distributed under the License is distributed on an "AS IS" BASIS,
#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#  See the License for the specific language governing permissions and
#  limitations under the License.
"""Example of using convolutional networks over characters for DBpedia dataset.

This model is similar to one described in this paper:
   "Character-level Convolutional Networks for Text Classification"
   http://arxiv.org/abs/1509.01626

and is somewhat alternative to the Lua code from here:
   https://github.com/zhangxiangxiao/Crepe
"""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import argparse
import sys

import numpy as np
import pandas
import tensorflow as tf

FLAGS = None

MAX_DOCUMENT_LENGTH = 100
N_FILTERS = 10
FILTER_SHAPE1 = [20, 256]
FILTER_SHAPE2 = [20, N_FILTERS]
POOLING_WINDOW = 4
POOLING_STRIDE = 2
MAX_LABEL = 15
CHARS_FEATURE = 'chars'  # Name of the input character feature.


def char_cnn_model(features, labels, mode):
  """Character level convolutional neural network model to predict classes."""
  features_onehot = tf.one_hot(features[CHARS_FEATURE], 256)
  input_layer = tf.reshape(
      features_onehot, [-1, MAX_DOCUMENT_LENGTH, 256, 1])
  with tf.variable_scope('CNN_Layer1'):
    # Apply Convolution filtering on input sequence.
    conv1 = tf.layers.conv2d(
        input_layer,
        filters=N_FILTERS,
        kernel_size=FILTER_SHAPE1,
        padding='VALID',
        # Add a ReLU for non linearity.
        activation=tf.nn.relu)
    # Max pooling across output of Convolution+Relu.
    pool1 = tf.layers.max_pooling2d(
        conv1,
        pool_size=POOLING_WINDOW,
        strides=POOLING_STRIDE,
        padding='SAME')
    # Transpose matrix so that n_filters from convolution becomes width.
    pool1 = tf.transpose(pool1, [0, 1, 3, 2])
  with tf.variable_scope('CNN_Layer2'):
    # Second level of convolution filtering.
    conv2 = tf.layers.conv2d(
        pool1,
        filters=N_FILTERS,
        kernel_size=FILTER_SHAPE2,
        padding='VALID')
    # Max across each filter to get useful features for classification.
    pool2 = tf.squeeze(tf.reduce_max(conv2, 1), axis=[1])

  # Apply regular WX + B and classification.
  logits = tf.layers.dense(pool2, MAX_LABEL, activation=None)

  predicted_classes = tf.argmax(logits, 1)
  if mode == tf.estimator.ModeKeys.PREDICT:
    return tf.estimator.EstimatorSpec(
        mode=mode,
        predictions={
            'class': predicted_classes,
            'prob': tf.nn.softmax(logits)
        })

  loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
  if mode == tf.estimator.ModeKeys.TRAIN:
    optimizer = tf.train.AdamOptimizer(learning_rate=0.01)
    train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step())
    return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op)

  eval_metric_ops = {
      'accuracy': tf.metrics.accuracy(
          labels=labels, predictions=predicted_classes)
  }
  return tf.estimator.EstimatorSpec(
      mode=mode, loss=loss, eval_metric_ops=eval_metric_ops)


def main(unused_argv):
  tf.logging.set_verbosity(tf.logging.INFO)

  # Prepare training and testing data
  dbpedia = tf.contrib.learn.datasets.load_dataset(
      'dbpedia', test_with_fake_data=FLAGS.test_with_fake_data, size='large')
  x_train = pandas.DataFrame(dbpedia.train.data)[1]
  y_train = pandas.Series(dbpedia.train.target)
  x_test = pandas.DataFrame(dbpedia.test.data)[1]
  y_test = pandas.Series(dbpedia.test.target)

  # Process vocabulary
  char_processor = tf.contrib.learn.preprocessing.ByteProcessor(
      MAX_DOCUMENT_LENGTH)
  x_train = np.array(list(char_processor.fit_transform(x_train)))
  x_test = np.array(list(char_processor.transform(x_test)))

  x_train = x_train.reshape([-1, MAX_DOCUMENT_LENGTH, 1, 1])
  x_test = x_test.reshape([-1, MAX_DOCUMENT_LENGTH, 1, 1])

  # Build model
  classifier = tf.estimator.Estimator(model_fn=char_cnn_model)

  # Train.
  train_input_fn = tf.estimator.inputs.numpy_input_fn(
      x={CHARS_FEATURE: x_train},
      y=y_train,
      batch_size=128,
      num_epochs=None,
      shuffle=True)
  classifier.train(input_fn=train_input_fn, steps=100)

  # Predict.
  test_input_fn = tf.estimator.inputs.numpy_input_fn(
      x={CHARS_FEATURE: x_test},
      y=y_test,
      num_epochs=1,
      shuffle=False)
  predictions = classifier.predict(input_fn=test_input_fn)
  y_predicted = np.array(list(p['class'] for p in predictions))
  y_predicted = y_predicted.reshape(np.array(y_test).shape)

  # Score with tensorflow.
  scores = classifier.evaluate(input_fn=test_input_fn)
  print('Accuracy: {0:f}'.format(scores['accuracy']))


if __name__ == '__main__':
  parser = argparse.ArgumentParser()
  parser.add_argument(
      '--test_with_fake_data',
      default=False,
      help='Test the example code with fake data.',
      action='store_true')
  FLAGS, unparsed = parser.parse_known_args()
  tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)