# -*- coding: utf-8 -*-

# ---

# jupyter:

# jupytext:

# formats: ipynb,py

# text_representation:

# extension: .py

# format_name: light

# format_version: '1.4'

# jupytext_version: 1.1.0

# kernelspec:

# display_name: Python 3

# language: python

# name: python3

# ---

# + {"colab": {}, "colab_type": "code", "id": "dRdv0ml29PSa"}

from keras.layers import Dense, Activation, Input, Conv2D, LSTM, MaxPool2D, UpSampling2D

from sklearn.model_selection import train_test_split

from keras.callbacks import EarlyStopping

from keras.utils import to_categorical

from numpy import argmax, array_equal

import matplotlib.pyplot as plt

from keras.models import Model

from random import randint

import pandas as pd

import numpy as np

# + {"colab": {}, "colab_type": "code", "id": "nA9_qW1_9WOo"}

train = pd.read_csv('./data/fashion-mnist_train.csv')

train_x = train[list(train.columns)[1:]].values

train_y = train['label'].values

# -

# ### データの確認

train.head()

# ### データセットの前処理

# - データセットを読み込み、訓練データと予測用のデータに分割

# - 正規化

# +

train_x = train_x / 255

train_x, val_x, train_y, val_y = train_test_split(train_x, train_y, test_size=0.2)

train_x = train_x.reshape(-1, 784)

val_x = val_x.reshape(-1, 784)

# + {"colab_type": "text", "id": "DOICGWUx1LQo", "cell_type": "markdown"}

# ### オートエンコーダーのアーキテクチャを作成

# - node数が2000, 12000, 500の3つの層で構成

# + {"colab": {}, "colab_type": "code", "id": "OXSfU2iH9ZbZ"}

input_layer = Input(shape=(784,))

encode_layer1 = Dense(1500, activation='relu')(input_layer)

encode_layer2 = Dense(1000, activation='relu')(encode_layer1)

encode_layer3 = Dense(500, activation='relu')(encode_layer2)

latent_view = Dense(10, activation='sigmoid')(encode_layer3)

decode_layer1 = Dense(500, activation='relu')(latent_view)

decode_layer2 = Dense(1000, activation='relu')(decode_layer1)

decode_layer3 = Dense(1500, activation='relu')(decode_layer2)

output_layer = Dense(784)(decode_layer3)

model = Model(input_layer, output_layer)

# -

# ### モデルのアーキテクチャを可視化

# +

import sys

sys.path.append('../../python_lib/convnet-drawer')

# +

from keras.models import Sequential

from convnet_drawer import Model, Conv2D, MaxPooling2D, Flatten, Dense

drawer_model = Model(input_shape=(28, 28, 3))

drawer_model.add(Flatten())

drawer_model.add(Dense(1500))

drawer_model.add(Dense(1000))

drawer_model.add(Dense(500))

drawer_model.add(Dense(10))

drawer_model.add(Dense(500))

drawer_model.add(Dense(1000))

drawer_model.add(Dense(1500))

drawer_model.add(Dense(784))

drawer_model.save_fig('alexnet.svg')

# + {"colab_type": "text", "id": "mvFdHRtk2GGY", "cell_type": "markdown"}

# ### モデルの概要を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 442}, "colab_type": "code", "id": "cr3WlpfZBd7a", "outputId": "75a74782-6a78-4ed2-bbb1-34dc3d770b83"}

model.summary()

# + {"colab_type": "text", "id": "t9yTfVZY2O01", "cell_type": "markdown"}

# ### モデルのコンパイル

# - EarlyStoppingの設定

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 731}, "colab_type": "code", "id": "Nv1BnA-39Z63", "outputId": "f9f6583b-e3ab-49da-f7e0-2a76d77488a5"}

model.compile(optimizer='adam', loss='mse')

early_stopping = EarlyStopping(monitor='val_loss', min_delta=0, patience=10, verbose=1, mode='auto')

history = model.fit(train_x, train_x, epochs=3, batch_size=2048, validation_data=(val_x, val_x), callbacks=[early_stopping])

# -

# ### 学習結果の可視化

# +

import matplotlib.pyplot as plt

# %matplotlib inline

#loss

plt.plot(history.history['loss'])

plt.plot(history.history['val_loss'])

plt.title('model loss')

plt.ylabel('loss')

plt.xlabel('epoch')

plt.legend(['train', 'test'], loc='upper left')

plt.show()

# -

preds = model.predict(val_x)

preds

# + {"colab_type": "text", "id": "_zLwSOGC3OO6", "cell_type": "markdown"}

# ### 入力画像を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 923}, "colab_type": "code", "id": "05dZ5nRyC1TO", "outputId": "a6998a6a-c28d-46ab-9d56-4ccf2cd89458"}

from PIL import Image

f, ax = plt.subplots(1, 5)

f.set_size_inches(80, 40)

for i in range(5):

ax[i].imshow(val_x[i].reshape(28, 28))

plt.show()

# + {"colab_type": "text", "id": "mC1Em-XT5Czu", "cell_type": "markdown"}

# ### 出力画像を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 838}, "colab_type": "code", "id": "jRXMDooXDRku", "outputId": "8430089e-69a0-44d8-abe9-145a1700e389"}

f, ax = plt.subplots(1, 5)

f.set_size_inches(80, 40)

for i in range(5):

ax[i].imshow(preds[i].reshape(28, 28))

plt.show()

# + {"colab_type": "text", "id": "z3Lf3MiK508h", "cell_type": "markdown"}

# ### 画像のノイズ除去

# + {"colab_type": "text", "id": "MdFfo8095b9z", "cell_type": "markdown"}

#

# + {"colab": {}, "colab_type": "code", "id": "G2V02lFtEXqV"}

train_x = train[list(train.columns)[1:]].values

train_x, val_x = train_test_split(train_x, test_size=0.2)

## normalize and reshape

train_x = train_x/255.

val_x = val_x/255.

# + {"colab": {}, "colab_type": "code", "id": "yom_kL8VwUw0"}

train_x = train_x.reshape(-1, 28, 28, 1)

val_x = val_x.reshape(-1, 28, 28, 1)

# + {"colab_type": "text", "id": "8BR7leP76D4g", "cell_type": "markdown"}

# ### ノイズを付与して、画像認識の精度を上げる

# - https://qiita.com/bohemian916/items/9630661cd5292240f8c7

# + {"colab": {}, "colab_type": "code", "id": "aQZ-668PwUzM"}

noise = augmenters.SaltAndPepper(0.1)

seq_object = augmenters.Sequential([noise])

train_x_n = seq_object.augment_images(train_x * 255) / 255

val_x_n = seq_object.augment_images(val_x * 255) / 255

# + {"colab_type": "text", "id": "44444jTc6bGi", "cell_type": "markdown"}

# ### 加工前の画像を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 838}, "colab_type": "code", "id": "KGuNwobroAW4", "outputId": "07dc019e-c36a-411c-ca3b-10d071eb4b50"}

f, ax = plt.subplots(1,5)

f.set_size_inches(80, 40)

for i in range(5,10):

ax[i-5].imshow(train_x[i].reshape(28,28))

plt.show()

# + {"colab_type": "text", "id": "VP9a2jpY6f4O", "cell_type": "markdown"}

# ### ノイズ付与後の画像を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 838}, "colab_type": "code", "id": "uBDH7meYojhm", "outputId": "1118f7ac-a8a7-4962-ae6f-3b0003142261"}

f, ax = plt.subplots(1,5)

f.set_size_inches(80, 40)

for i in range(5,10):

ax[i-5].imshow(train_x_n[i].reshape(28,28))

plt.show()

# + {"colab_type": "text", "id": "9SoVatND6kdx", "cell_type": "markdown"}

# ### プーリング層を追加してモデル構築

# + {"colab": {}, "colab_type": "code", "id": "UkX3hoNaFJLt"}

input_layer = Input(shape=(28, 28, 1))

encoded_layer1 = Conv2D(64, (3,3), activation='relu', padding='same')(input_layer)

encoded_layer1 = MaxPool2D((2,2), padding='same')(encoded_layer1)

encoded_layer2 = Conv2D(32, (3,3), activation='relu', padding='same')(encoded_layer1)

encoded_layer2 = MaxPool2D((2,2), padding='same')(encoded_layer2)

encoded_layer3 = Conv2D(16, (3,3), activation='relu', padding='same')(encoded_layer2)

latent_view = MaxPool2D((2,2), padding='same')(encoded_layer3)

decoded_layer1 = Conv2D(16, (3,3), activation='relu', padding='same')(latent_view)

decoded_layer1 = UpSampling2D((2,2))(decoded_layer1)

decoded_layer2 = Conv2D(32, (3,3), activation='relu', padding='same')(decoded_layer1)

decoded_layer2 = UpSampling2D((2,2))(decoded_layer2)

decoded_layer3 = Conv2D(64, (3,3), activation='relu')(decoded_layer2)

decoded_layer3 = UpSampling2D((2,2))(decoded_layer3)

output_layer = Conv2D(1, (3,3), padding='same')(decoded_layer3)

model_2 = Model(input_layer, output_layer)

model_2.compile(optimizer='adam', loss='mse')

# + {"colab_type": "text", "id": "fHH62KrI7Um2", "cell_type": "markdown"}

# ### モデルの概要を表示

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 629}, "colab_type": "code", "id": "Kob51f3Lpjrk", "outputId": "9daf8bfa-5e2b-4930-b6ea-d5b5ce478503"}

model_2.summary()

# + {"colab_type": "text", "id": "utPawuva7e6X", "cell_type": "markdown"}

# ### EarlyStoppingの設定

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 374}, "colab_type": "code", "id": "qgfNMei5ftH1", "outputId": "eb8986b1-c026-463a-981f-bb33848b62a3"}

early_stopping = EarlyStopping(monitor='val_loss', min_delta=0, patience=10, verbose=5, mode='auto')

history = model_2.fit(train_x_n, train_x, epochs=10, batch_size=2048, validation_data=(val_x_n, val_x), callbacks=[early_stopping])

# + {"colab": {"base_uri": "https://localhost:8080/", "height": 838}, "colab_type": "code", "id": "_OZPx2kcinZi", "outputId": "a54ee176-3daa-4a27-b185-b8c393697458"}

preds = model_2.predict(val_x_n[:10])

f, ax = plt.subplots(1,5)

f.set_size_inches(80, 40)

for i in range(5,10):

ax[i-5].imshow(preds[i].reshape(28, 28))

plt.show()

# + {"colab_type": "text", "id": "7o963kGV78SK", "cell_type": "markdown"}

# ### オートエンコーダを用いたシーケンス間予測

# + {"colab": {}, "colab_type": "code", "id": "zlMOCK-EqVhm"}

def dataset_preparation(n_in, n_out, n_unique, n_samples):

inputs[0][0]

# + {"colab": {}, "colab_type": "code", "id": "f-hp0rkK0URf"}