Kerasによるモデル作成¶
- MLP（多層パーセプトロン）
- CNN（畳み込みニューラルネットワーク）
のモデル作成方法の習得。
※ここでは、ハイパーパラメータチューニングは省略
Kerasでアヤメのデータ分類¶
In [ ]:
# 上記モデリングをKerasでコーディングした場合が以下。
import tensorflow as tf
import tensorflow.contrib.keras as keras
import pandas as pd
import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
In [ ]:
# アヤメデータ読み込み
iris = load_iris()
# ターゲット変数をone-hotベクトル化
target_one_hot = pd.DataFrame({'iris': iris.target})
target_one_hot = pd.get_dummies(target_one_hot['iris'])
# one-hotベクトル化したターゲット変数をnp.array化
y_nums = target_one_hot.to_numpy()
y_nums[:5]
Out[ ]:
In [ ]:
# 学習データの用意と分割
# 学習データ
x_data = iris.data
# 訓練とテスト用に分割
x_train, x_test, y_train, y_test = train_test_split(x_data, y_nums, test_size=0.2, random_state=8)
In [ ]:
# モデル定義
Dense = keras.layers.Dense
model = keras.models.Sequential()
model.add(Dense(10, activation='relu', input_shape=(4,)))
model.add(Dense(3, activation='softmax'))
# モデル構築
model.compile(
loss = 'categorical_crossentropy',
optimizer = 'adam',
metrics = ['accuracy']
)
In [ ]:
# 学習実行
model.fit(x_train, y_train, batch_size=20, epochs=300)
Out[ ]:
In [ ]:
# モデルの評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", str(score[1]), "loss＝",score[0])
In [ ]:
score
Out[ ]:
手書き数字の判定¶
- kearasは、tensorflow2.2以上が必要なので、tensorflowのバージョンを最新にする
In [ ]:
# !pip install --upgrade tensorflow
In [ ]:
import tensorflow as tf
import keras
from keras.datasets import mnist
import pandas as pd
import numpy as np
from matplotlib import pyplot
In [ ]:
print(tf.__version__)
In [ ]:
# MNISTのデータ読み込み
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# データを４*８で出力
for i in range(32):
pyplot.subplot(4, 8, i+1)
pyplot.imshow(x_train[i], cmap='gray')
pyplot.show()
In [ ]:
x_train[0].shape
Out[ ]:
In [ ]:
np.max(x_train[0])
Out[ ]:
In [ ]:
# 画像１つが二次元なので、２８×２８で７８４の一次元配列に変換。
# また、データの範囲を0.0~1.0の範囲に正規化する必要があるので、色の最大値２５５で割る
# 学習データを28*28=784の一次元配列に変換し、正規化
x_train = x_train.reshape(-1, 784).astype('float32') / 255
x_test= x_test.reshape(-1, 784).astype('float32') / 255
In [ ]:
print(x_train[0].shape)
print(np.max(x_train[0]))
In [ ]:
x_test
Out[ ]:
In [ ]:
# 目的変数をone-hotベクトル化
print("目的変数：", y_train[:10], "・・・")
# kerasでone-hot化
y_train = keras.utils.to_categorical(y_train.astype('int32'), 10)
y_test = keras.utils.to_categorical(y_test.astype('int32'), 10)
In [ ]:
y_train[:2]
Out[ ]:
In [ ]:
y_test[:2]
Out[ ]:
In [ ]:
# kerasでモデル構築
# 入力と出力を指定
in_size = 28 * 28
out_size = 10
# モデル定義
Dense = keras.layers.Dense
model = keras.models.Sequential()
model.add(Dense(512, activation='relu', input_shape=(in_size,)))
model.add(Dense(out_size, activation='softmax'))
In [ ]:
# モデル構築
model.compile(
loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
In [ ]:
# 学習
model.fit(x_train, y_train, batch_size=20, epochs=20)
Out[ ]:
In [ ]:
# 評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", score[1], 'loss＝', score[0])
MLPでMNISTの分類¶
- 多層パーセプトロン(MLP)の利用
In [ ]:
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.optimizers import RMSprop
from keras.datasets import mnist
import matplotlib.pyplot as plt
In [ ]:
# 入力と出力の指定
in_size = 28 * 28
out_size = 10
# MNISTのデータ読み込み
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# データを28*28=784の一次元配列に変換し、正規化(最小0、最大1)
x_train = x_train.reshape(-1, 784).astype('float32') / 255
x_test = x_test.reshape(-1, 784).astype('float32') / 255
# ターゲット変数をone-hotベクトル化
y_train = keras.utils.to_categorical(y_train.astype('int32'), 10)
y_test = keras.utils.to_categorical(y_test.astype('int32'), 10)
In [ ]:
# MLPモデルの定義
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(in_size,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(out_size, activation='softmax'))
In [ ]:
# モデルのコンパイル
model.compile(
loss='categorical_crossentropy',
optimizer='RMSprop',
metrics=['accuracy']
)
In [ ]:
# 学習
hist = model.fit(
x_train, y_train,
batch_size=128,
epochs=50,
verbose=1,
validation_data=(x_test, y_test)
)
In [ ]:
# モデルの評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", score[1], "loss＝", score[0])
In [ ]:
# 学習の様子をグラフ化
# 正解率の推移
plt.plot(hist.history['accuracy'])
plt.plot(hist.history['val_accuracy'])
plt.title('Accuracy')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
In [ ]:
# ロスの推移
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('Loss')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
CNN(畳み込みニューラルネットワーク)の適用¶
- 上記の予測モデルをCNNで作成
In [ ]:
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras.optimizers import RMSprop
from keras.datasets import mnist
import matplotlib.pyplot as plt
In [ ]:
# 入力と出力を指定
im_rows = 28 # 画像の縦ピクセルサイズ
im_cols = 28 # 画像の横ピクセルサイズ
im_color = 1 # 画像の色空間／グレイスケール
in_shape = (im_rows, im_cols, im_color) # 入力サイズ
out_size = 10 # 出力サイズ（１行１０列）
In [ ]:
# MNISTのデータ読み込み
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train[0].shape
Out[ ]:
In [ ]:
x_train.shape
Out[ ]:
In [ ]:
y_train.shape
Out[ ]:
In [ ]:
# 読み込んだ各データを2次元から3次元配列に変換後、正規化
# （今までは、各画像データを1次元に変換していたが、ここでは、3次元に変換する）
x_train = x_train.reshape(-1, im_rows, im_cols, im_color)
x_train = x_train.astype('float32') / 255
x_test = x_test.reshape(-1, im_rows, im_cols, im_color)
x_test = x_test.astype('float32') / 255
In [ ]:
x_train.shape
Out[ ]:
In [ ]:
x_train[0][:2]
Out[ ]:
In [ ]:
# ターゲット変数をone-hotベクトル化
y_train = keras.utils.to_categorical(y_train.astype('int32'), 10)
y_test = keras.utils.to_categorical(y_test.astype('int32'), 10)
In [ ]:
# CNNモデルを定義
model = Sequential()
model.add(Conv2D(
32,
kernel_size=(3, 3),
activation='relu',
input_shape=in_shape
))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(out_size, activation='softmax'))
In [ ]:
# モデルのコンパイル
model.compile(
loss='categorical_crossentropy',
optimizer=RMSprop(),
metrics=['accuracy']
)
In [ ]:
# 学習
hist = model.fit(
x_train, y_train,
batch_size=128,
epochs=12,
verbose=1,
validation_data=(x_test, y_test)
)
In [ ]:
# モデルの評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", score[1], "loss＝", score[0])
In [ ]:
# 学習推移のプロット
# 正解率
plt.plot(hist.history['accuracy'])
plt.plot(hist.history['val_accuracy'])
plt.title('Accuracy')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
In [ ]:
# ロスの推移
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('Loss')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
写真の物体認識¶
In [ ]:
# 対象写真データ確認
from keras.datasets import cifar10
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
In [ ]:
from PIL import Image
In [ ]:
plt.figure(figsize=(10, 10))
labels = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
for i in range(40):
im = Image.fromarray(x_train[i])
plt.subplot(5, 8, i+1)
plt.title(labels[y_train[i][0]])
plt.tick_params(labelbottom="off", bottom="off") # x軸をoff
plt.tick_params(labelleft="off", left="off") # y軸をoff
plt.imshow(im)
plt.show()
In [ ]:
x_train.shape
Out[ ]:
In [ ]:
x_train[0]
Out[ ]:
実装¶
In [ ]:
import matplotlib.pyplot as plt
import keras
from keras.datasets import cifar10
from keras.models import Sequential
from keras.layers import Dense, Dropout
In [ ]:
num_classes = 10
im_rows = 32
im_cols = 32
im_size = im_rows * im_cols * 3
In [ ]:
# データ読み込み
(x_train, y_train), (x_test, y_test) =cifar10.load_data()
In [ ]:
# データを１次元配列に変換後、正規化
x_train = x_train.reshape(-1, im_size).astype('float32') / 255
x_test = x_test.reshape(-1, im_size).astype('float32') / 255
# ターゲット変数をone-hotベクトル化
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
In [ ]:
x_train[0]
Out[ ]:
In [ ]:
y_train
Out[ ]:
In [ ]:
y_train.shape
Out[ ]:
In [ ]:
# モデルの定義
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(im_size,)))
model.add(Dense(num_classes, activation='softmax'))
In [ ]:
# モデルのコンパイル
model.compile(
loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
In [ ]:
# 学習
hist = model.fit(
x_train, y_train,
batch_size=32,
epochs=50,
verbose=1,
validation_data=(x_test, y_test)
)
In [ ]:
# 評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", score[1], "loss＝", score[0])
In [ ]:
# 学習推移の可視化
# Accuracy
plt.plot(hist.history['accuracy'])
plt.plot(hist.history['val_accuracy'])
plt.title("Accuracy")
plt.legend(['train', 'test'], loc='upper left')
plt.show()
In [ ]:
# loss
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title("Loss")
plt.legend(['train', 'test'], loc='upper left')
plt.show()
上記の予測をCNNで実施¶
In [ ]:
import matplotlib.pyplot as plt
import keras
from keras.datasets import cifar10
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Conv2D, MaxPooling2D
In [ ]:
# 入力データと出力データの次元数セット
num_classes = 10
im_rows = 32
im_cols = 32
in_shape = (im_rows, im_cols, 3)
In [ ]:
# データ読み込み
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
In [ ]:
print(x_train.shape)
print(x_test.shape)
In [ ]:
# 学習データの正規化
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255
# ターゲット変数をone-hotベクトル化
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
In [ ]:
# モデル定義
model = Sequential()
model.add(Conv2D(32, (3,3), padding='same', input_shape=in_shape))
model.add(Activation('relu'))
model.add(Conv2D(32, (3,3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))
model.add(Conv2D(64, (3,3), padding='same'))
model.add(Activation('relu'))
model.add(Conv2D(64, (3,3)))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes))
model.add(Activation('softmax'))
In [ ]:
# モデルのコンパイル
model.compile(
loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
In [ ]:
# 学習
hist = model.fit(
x_train, y_train,
batch_size=32,
epochs=50,
verbose=1,
validation_data=(x_test, y_test)
)
In [ ]:
# 評価
score = model.evaluate(x_test, y_test, verbose=1)
print("正解率＝", score[1], "loss＝", score[0])
In [ ]:
# 学習推移の可視化
# Accuracy
plt.plot(hist.history['accuracy'])
plt.plot(hist.history['val_accuracy'])
plt.title("Accuracy")
plt.legend(['train', 'test'], loc='upper left')
plt.show()
In [ ]:
# loss
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title("Loss")
plt.legend(['train', 'test'], loc='upper left')
plt.show()
In [ ]:
model.save_weights('/content/drive/My Drive/画像認識/DL_data/cifar10-weight.h5')
In [ ]:
x_test[0].shape
Out[ ]:
上記モデルによる写真の予測¶
In [ ]:
import cv2
import numpy as np
In [ ]:
labels = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
im_size = 32 * 32 * 3
In [ ]:
# モデルデータ読み込み
model.load_weights('/content/drive/My Drive/画像認識/DL_data/cifar10-weight.h5')
In [ ]:
# openCVで画像読み込み
im = cv2.imread('/content/drive/My Drive/画像認識/DL_data/cat.jpg')
# 色空間を変換して、リサイズ
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
im = cv2.resize(im, (32, 32))
plt.imshow(im)
plt.show()
In [ ]:
# MLPで学習した画像データに合わせる
im = im.astype('float32') / 255
# 予測
r = model.predict(np.array([im]), batch_size=32, verbose=1)
res = r[0]
In [ ]:
# 結果
for i, acc in enumerate(res):
print(labels[i], '=', int(acc * 100))
print('---')
print('予測した結果 = ', labels[res.argmax()])
In [ ]: