[TensorFlow Certification Day6] 第二張上課證照入手

Explore multi-class with Rock Paper Scissors dataset (CNN/Week4) 3hrs

主要利用剪刀石頭布的data去訓練多分類的題目
關鍵在下面這段, 把binary 變成categorical, 還有loss也改成categorical_crossentropy, 最後一層改成softmax

train_datagen = ImageDataGenerator(rescale=1/255)

#'binary' 變成 'categorical'
train_generator = train_datagen.flow_from_directory(
            train_dir,
            target_size=(300, 300),
            batch_size=128,
            class_mode='categorical')

#把output layer變成softmax
tf.keras.layers.Dense(3, activation='softmax')

#loss function改成'categorical_crossentropy'
model.compile(loss='categorical_crossentroy',
                            optimizer=RMSprop(lr=0.001),
                            metrics=['acc'])

但是如果真的以為只有這樣那就錯了, Quiz還是一樣不難, 但是exercise難度卻開始往上加
這邊把遇到的坑做一個筆記

#在get data要學習怎麼處理csv -> data
def get_data(filename):
    images = []
    labels = []
    with open(filename) as training_file:
        csvreader = csv.reader(training_file)
        next(csvreader)
        for row in csvreader:
            images.append(np.array_split(row[1:],28))
            temparr = np.zeros(10, int)
            labels.append(int(row[0]))
      # Your code starts here
      # Your code ends here
    images = np.asarray(images, dtype=float)
    labels = np.asarray(labels, dtype=float)

    return images, labels

#課堂學到的softmax, 多分類
model = tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(16, (3,3), activation='relu', input_shape=(28, 28, 1)),
    tf.keras.layers.MaxPooling2D(2, 2),
    # The second convolution
    tf.keras.layers.Conv2D(32, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),

    # Flatten the results to feed into a DNN
    tf.keras.layers.Flatten(),
    # 512 neuron hidden layer
    tf.keras.layers.Dense(512, activation='relu'),
    # Only 1 output neuron. It will contain a value from 0-1 where 0 for 1 class ('horses') and 1 for the other ('humans')
    tf.keras.layers.Dense(25, activation='softmax')
])

#採用sparse_categorical_crossentropy就可以不用做one-hot encode
model.compile(loss='sparse_categorical_crossentropy',
             optimizer='adam',
             metrics=['acc'])

#最後是ImageDataGenerator因為這次不是使用directory直接餵的方式, 所以直接在fit_generator裡面把
#datagen放進來, 並使用.flow function, 帶入images, labels, batch_size就大功告成
history = model.fit_generator(
    train_datagen.flow(training_images, training_labels, batch_size=8),
    steps_per_epoch=2000,
    epochs=2
)

這樣子總算是把第二章上課證拿到了, 做到這裡會發現, 其實API不難處理
其實難處理的是data怎麼拿出來跟怎麼fit 這些API接口
這裡建議要多練習, 或是做好準備, 否到時候會debug半天