tf.keras Build neural network function expansion

One 、 Homemade data sets , Solve the application problems in this field

1.1 Observation data set data structure , Match as a feature tag pair

mnist_image_label Folder ：

Picture folder ：

Label folder ：

def generateds(path, txt):
    f = open(txt, 'r')  #  Open as read-only txt file 
    contents = f.readlines()  #  Read all lines in the file 
    f.close()  #  close txt file 
    x, y_ = [], []  #  Create an empty list 
    for content in contents:  #  Line by line 
        value = content.split()  #  Separate... By spaces , The picture path is value[0] ,  The label is value[1] ,  Save list 
        img_path = path + value[0]  #  Spell out the picture path and file name 
        img = Image.open(img_path)  #  Read in the picture 
        img = np.array(img.convert('L'))  #  The picture becomes 8 Bit width gray value np.array Format 
        img = img / 255.  #  Data normalization  （ Implement preprocessing ）
        x.append(img)  #  Normalized data , Paste to list x
        y_.append(value[1])  #  Label to list y_
        print('loading : ' + content)  #  Print status prompt 

    x = np.array(x)  #  Turn into np.array Format 
    y_ = np.array(y_)  #  Turn into np.array Format 
    y_ = y_.astype(np.int64)  #  Turn into 64 An integer 
    return x, y_  #  Returns the input feature x, Return label y_

print('-------------Generate Datasets-----------------')
    x_train, y_train = generateds(train_path, train_txt)
    x_test, y_test = generateds(test_path, test_txt)

Two 、 Data to enhance , Increase the amount of data

2.1 Data to enhance

image_gen_train=tf.keras.preprocessing.image.ImageDataGenerator( Enhancement method )
image_gen_train.fit(x_train)

Common enhancement methods ：

• Scaling factor ：rescale= All data will be multiplied by the value provided
• Random rotation ：rotation_range= Range of degrees of random rotation angle
• Width offset ：width_shift_range= Random width offset
• Height offset ：height_shift_range= Random height offset
• Flip horizontal ：horizontal_flip= Whether to flip horizontally and randomly
• Random scaling ：zoom_range= Range of random scaling [1-n,1+n]

import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)  #  Add a dimension to the data , from (60000, 28, 28)reshape by (60000, 28, 28, 1)

image_gen_train = ImageDataGenerator(
    rescale=1. / 1.,  #  If it is an image , The denominator is 255 when , Attributable to 0～1
    rotation_range=45,  #  Random 45 Degree of rotation 
    width_shift_range=.15,  #  Width offset 
    height_shift_range=.15,  #  Height offset 
    horizontal_flip=False,  #  Flip horizontal 
    zoom_range=0.5  #  Randomly scale the image by the threshold amount 50％
)
image_gen_train.fit(x_train)

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])

model.fit(image_gen_train.flow(x_train, y_train, batch_size=32), epochs=5, validation_data=(x_test, y_test),
          validation_freq=1)
model.summary()

2.2 Data enhanced visualization

3、 ... and 、 Break point continued training , Access model

3.1 Read the model

load_weights( Pathfilename )

checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
    print('-------------load the model-----------------')
    model.load_weights(checkpoint_save_path)

3.2 Save the model

With the help of tensorflow The callback function given , Save parameters and network directly

tf.keras.callbacks.ModelCheckpoint(
    filepath= Pathfilename , 
    save_weights_only=True, 
    monitor='val_loss', # val_loss or loss 
    save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, 
    validation_data=(x_test, y_test), validation_freq=1, 
    callbacks=[cp_callback])
)

import tensorflow as tf
import os

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])

checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
    print('-------------load the model-----------------')
    model.load_weights(checkpoint_save_path)

cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
                                                 save_weights_only=True,
                                                 save_best_only=True)

history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
                    callbacks=[cp_callback])
model.summary()

Four 、 Parameter extraction , Write to text

4.1 Extract trainable parameters

model.trainable_variables Trainable parameters in the model

4.2 Set up print Output format

np.set_printoptions(precision= The decimal point shall be rounded to several places ,threshold= The number of array elements is less than or equal to the threshold , Print all elements ; Otherwise, print the threshold +1 Elements , Use an ellipsis in the middle to supplement )

>>> np.set_printoptions(precision=5)
>>> print(np.array([1.123456789]))
[1.12346]
>>> np.set_printoptions(threshold=5)
>>> print(np.arange(10))
[0 1 2 … , 7 8 9]

notes ：precision=np.inf Print full decimal places ;threshold=np.nan Print all array elements .

import tensorflow as tf
import os
import numpy as np
np.set_printoptions(threshold=np.inf)
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])
checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
    print('-------------load the model-----------------')
    model.load_weights(checkpoint_save_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
                                                 save_weights_only=True,
                                                 save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
                    callbacks=[cp_callback])
model.summary()
print(model.trainable_variables)
file = open('./weights.txt', 'w')
for v in model.trainable_variables:
    file.write(str(v.name) + '\n')
    file.write(str(v.shape) + '\n')
    file.write(str(v.numpy()) + '\n')
file.close()

weights.txt：

5、 ... and 、acc/loss visualization , See how your workout works

history=model.fit( Training set data ,  Training set label , batch_size=, epochs=,
    validation_split= Proportion used as test data ,validation_data= Test set , validation_freq= Test frequency )

history：
loss： Training set loss
val_loss： Test set loss
sparse_categorical_accuracy： Training set accuracy
val_sparse_categorical_accuracy： Test set accuracy

acc = history.history['sparse_categorical_accuracy']
val_acc = history.history['val_sparse_categorical_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']

plt.subplot(1, 2, 1)
plt.plot(acc, label='Training Accuracy')
plt.plot(val_acc, label='Validation Accuracy')
plt.title('Training and Validation Accuracy')
plt.legend()

plt.subplot(1, 2, 2)
plt.plot(loss, label='Training Loss')
plt.plot(val_loss, label='Validation Loss')
plt.title('Training and Validation Loss')
plt.legend()
plt.show()

6、 ... and 、 Applications , Figure identification

from PIL import Image
import numpy as np
import tensorflow as tf

model_save_path = './checkpoint/mnist.ckpt'

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')])

model.load_weights(model_save_path)

preNum = int(input("input the number of test pictures:"))

for i in range(preNum):
    image_path = input("the path of test picture:")
    img = Image.open(image_path)
    img = img.resize((28, 28), Image.ANTIALIAS)
    img_arr = np.array(img.convert('L'))

    for i in range(28):
        for j in range(28):
            if img_arr[i][j] < 200:
                img_arr[i][j] = 255
            else:
                img_arr[i][j] = 0

    img_arr = img_arr / 255.0
    x_predict = img_arr[tf.newaxis, ...]
    result = model.predict(x_predict)

    pred = tf.argmax(result, axis=1)

    print('\n')
    tf.print(pred)