import numpy as np
import pandas as pd
import tensorflow as tf

TRAIN_URL = "http://download.tensorflow.org/data/iris_training.csv"
TEST_URL = "http://download.tensorflow.org/data/iris_test.csv"

CSV_COLUMN_NAMES = ['SepalLength', 'SepalWidth',
                    'PetalLength', 'PetalWidth', 'Species']
SPECIES = ['Sentosa', 'Versicolor', 'Virginica']

def maybe_download():
    """ iris dataset을 다운 받고 다운 받은 곳의 path값을 받아온다."""
    train_path = tf.keras.utils.get_file(TRAIN_URL.split('/')[-1], TRAIN_URL)
    test_path = tf.keras.utils.get_file(TEST_URL.split('/')[-1], TEST_URL)
    return train_path, test_path

def load_data(y_name='Species'):
    """Returns the iris dataset as (train_x, train_y), (test_x, test_y)."""
    train_path, test_path = maybe_download()

    train = pd.read_csv(train_path, names=CSV_COLUMN_NAMES, header=0)
    train_x, train_y = train, train.pop(y_name)

    test = pd.read_csv(test_path, names=CSV_COLUMN_NAMES, header=0)
    test_x, test_y = test, test.pop(y_name)

    return (train_x, train_y), (test_x, test_y)

(train_x, train_y), (test_x, test_y) = load_data()
print(np.shape(train_x))    # 결과 : (120, 4)
print(np.shape(train_y))    # 결과 : (120, )
print(np.shape(test_x))     # 결과 : (30, 4)
print(np.shape(test_y))     # 결과 : (30, )

train_x.head(5)

train_y.head(5)

test_x.head(5)

test_y.head(5)

def get_shuffle_input_pipeline(train_filename, batch_size):
    # iris train datasets 의 column size을 알아낸다.
    column_size = len(open(train_filename).read().split(','))

    # TextLineReader를 이용해 dataset을 읽어온다.
    filename_queue = tf.train.string_input_producer([train_filename])
    reader = tf.TextLineReader(skip_header_lines=1)
    _, values = reader.read(filename_queue)
    record_default = [[0.0] for _ in range(column_size)]
    feature = fields[:-1]
    label = fields[-1]

    # shuffle batch pipeline을 만들어준다.
    batch_feature, batch_label = tf.train.shuffle_batch(
        [feature, label],
        batch_size = batch_size,
        num_threads=4,
        min_after_dequeue=100,
        capacity=1000,
    )
    return batch_feature, batch_label


def train_fn(batch_set, epoch_size, batch_size, total_lines):
    """Model(DNN, CNN, RNN etc) Train"""
    batch_feature, batch_label = batch_set

    """ Model Graph section (모델 그래프 정의 구간)"""

    with tf.Session() as session:
        session.run(tf.global_variables_initializer())  # Session init
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(sess=session, coord=coord)

        one_epoch_size = int(total_lines / batch_size)
        for epoch in range(epoch_size):
            for _ in range(one_epoch_size):
                f_batch, l_batch = session.run([batch_feature, batch_label])
                """
                model train code
                """
                break
            break

        coord.request_stop()
        coord.join(threads)

epoch_size = 10  # 반복할 크기
batch_size = 4   # 배치 사이즈
train_file_path, test_file_path = maybe_download() # iris dataset을 다운 받고 다운 받은 곳의 path값을 받아온다.
total_dataset_size = len(open(train_file_path).readlines()) # 전체 데이터셋 사이즈

batch_graph = get_shuffle_input_pipeline(train_file_path, epoch_size, batch_size, total_dataset_size)
train_fn(batch_graph, epoch_size, batch_size, total_dataset_size)

def get_shuffle_input_pipeline(train_file_path, epoch_size, batch_size):
    frame = pd.read_csv(train_file_path, names=CSV_COLUMN_NAMES, header=0)
    features, labels = frame, frame.pop(CSV_COLUMN_NAMES[-1])
    dataset = tf.data.Dataset.from_tensor_slices((dict(features), labels))
    dataset = dataset.shuffle(1000).repeat(epoch_size).batch(batch_size) # 굉장히 직관적이고 쉽게 데이터 파이프 라인을 구성
    return dataset.make_one_shot_iterator().get_next() # one_shot 으로 한 방에

def train_fn(batch_set):
    """Model(DNN, CNN, RNN etc) Train"""

    """ Model Graph section (모델 그래프 정의 구간)"""
    with tf.Session() as session:
        session.run(tf.global_variables_initializer())
        while True:
            try:
                f_batch, l_batch = session.run(batch_set)
                """
                model train code
                """
            except Exception:
                break

epoch_size = 10 # 반복할 크기
batch_size = 4  # 배치 사이즈
train_file_path, test_file_path = maybe_download()
batch_set = get_shuffle_input_pipeline(train_file_path, epoch_size, batch_size)
train_fn(batch_set)