Tensorflow 1:mnist

Python
Author

Tony Duan

Published

October 10, 2023

1 about the data

The MNIST database (Modified National Institute of Standards and Technology database[1]) is a large database of handwritten digits(0-9) that is commonly used for training various image processing systems.

2 package

Code 
import numpy as np
import matplotlib.pyplot as plt
import os
import tensorflow as tf
print("TensorFlow version:", tf.__version__)
TensorFlow version: 2.14.0

3 input data

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

3.1 x_train:60000 pic with 28 by 28 pixel

Code 
x_train.shape
(60000, 28, 28)

3.2 x_test:10000 pic with 28 by 28 pixel

Code 
x_test.shape
(10000, 28, 28)

3.3 y_train:60000 pic lable

Code 
y_train.shape
(60000,)

almost each number have ~6000 so total 60,000 on trainning

Code 
unique, counts = np.unique(y_train, return_counts=True)
dict(zip(unique, counts))
{0: 5923,
 1: 6742,
 2: 5958,
 3: 6131,
 4: 5842,
 5: 5421,
 6: 5918,
 7: 6265,
 8: 5851,
 9: 5949}

3.4 y_train:10000 pic lable

Code 
y_test.shape
(10000,)

4 data clean

the pixel values of the images range from 0 through 255. Scale these values to a range of 0 to 1 by dividing the values by 255.0

Code 
x_train, x_test = x_train / 255.0, x_test / 255.0

5 bulid model

5.1 model

Code 
model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10)
])

5.2 loss function

Code 
loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

5.3 compile

Code 
model.compile(optimizer='adam',
              loss=loss_fn,
              metrics=['accuracy'])
Code 
# Show the model architecture
model.summary()
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 flatten (Flatten)           (None, 784)               0         
                                                                 
 dense (Dense)               (None, 128)               100480    
                                                                 
 dropout (Dropout)           (None, 128)               0         
                                                                 
 dense_1 (Dense)             (None, 10)                1290      
                                                                 
=================================================================
Total params: 101770 (397.54 KB)
Trainable params: 101770 (397.54 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

5.4 traning

Code 
model.fit(x_train, y_train, epochs=3)
Epoch 1/3
   1/1875 [..............................] - ETA: 3:57 - loss: 2.2271 - accuracy: 0.1250 100/1875 [>.............................] - ETA: 0s - loss: 0.9847 - accuracy: 0.7128   202/1875 [==>...........................] - ETA: 0s - loss: 0.7288 - accuracy: 0.7853 306/1875 [===>..........................] - ETA: 0s - loss: 0.6125 - accuracy: 0.8190 412/1875 [=====>........................] - ETA: 0s - loss: 0.5378 - accuracy: 0.8422 514/1875 [=======>......................] - ETA: 0s - loss: 0.4939 - accuracy: 0.8560 617/1875 [========>.....................] - ETA: 0s - loss: 0.4638 - accuracy: 0.8656 721/1875 [==========>...................] - ETA: 0s - loss: 0.4325 - accuracy: 0.8742 825/1875 [============>.................] - ETA: 0s - loss: 0.4108 - accuracy: 0.8804 927/1875 [=============>................] - ETA: 0s - loss: 0.3913 - accuracy: 0.88581031/1875 [===============>..............] - ETA: 0s - loss: 0.3746 - accuracy: 0.89051134/1875 [=================>............] - ETA: 0s - loss: 0.3619 - accuracy: 0.89431238/1875 [==================>...........] - ETA: 0s - loss: 0.3484 - accuracy: 0.89821341/1875 [====================>.........] - ETA: 0s - loss: 0.3362 - accuracy: 0.90181444/1875 [======================>.......] - ETA: 0s - loss: 0.3262 - accuracy: 0.90461546/1875 [=======================>......] - ETA: 0s - loss: 0.3173 - accuracy: 0.90711645/1875 [=========================>....] - ETA: 0s - loss: 0.3107 - accuracy: 0.90921749/1875 [==========================>...] - ETA: 0s - loss: 0.3035 - accuracy: 0.91141850/1875 [============================>.] - ETA: 0s - loss: 0.2971 - accuracy: 0.91311875/1875 [==============================] - 1s 490us/step - loss: 0.2954 - accuracy: 0.9136
Epoch 2/3
   1/1875 [..............................] - ETA: 1s - loss: 0.1279 - accuracy: 0.9375 107/1875 [>.............................] - ETA: 0s - loss: 0.1609 - accuracy: 0.9518 211/1875 [==>...........................] - ETA: 0s - loss: 0.1698 - accuracy: 0.9474 315/1875 [====>.........................] - ETA: 0s - loss: 0.1655 - accuracy: 0.9486 418/1875 [=====>........................] - ETA: 0s - loss: 0.1614 - accuracy: 0.9513 521/1875 [=======>......................] - ETA: 0s - loss: 0.1574 - accuracy: 0.9530 618/1875 [========>.....................] - ETA: 0s - loss: 0.1528 - accuracy: 0.9540 720/1875 [==========>...................] - ETA: 0s - loss: 0.1537 - accuracy: 0.9537 823/1875 [============>.................] - ETA: 0s - loss: 0.1516 - accuracy: 0.9545 926/1875 [=============>................] - ETA: 0s - loss: 0.1498 - accuracy: 0.95551028/1875 [===============>..............] - ETA: 0s - loss: 0.1481 - accuracy: 0.95631131/1875 [=================>............] - ETA: 0s - loss: 0.1467 - accuracy: 0.95671233/1875 [==================>...........] - ETA: 0s - loss: 0.1458 - accuracy: 0.95681336/1875 [====================>.........] - ETA: 0s - loss: 0.1445 - accuracy: 0.95761439/1875 [======================>.......] - ETA: 0s - loss: 0.1441 - accuracy: 0.95791541/1875 [=======================>......] - ETA: 0s - loss: 0.1439 - accuracy: 0.95791644/1875 [=========================>....] - ETA: 0s - loss: 0.1430 - accuracy: 0.95831747/1875 [==========================>...] - ETA: 0s - loss: 0.1425 - accuracy: 0.95841849/1875 [============================>.] - ETA: 0s - loss: 0.1421 - accuracy: 0.95841875/1875 [==============================] - 1s 489us/step - loss: 0.1419 - accuracy: 0.9585
Epoch 3/3
   1/1875 [..............................] - ETA: 1s - loss: 0.0101 - accuracy: 1.0000 104/1875 [>.............................] - ETA: 0s - loss: 0.1184 - accuracy: 0.9642 206/1875 [==>...........................] - ETA: 0s - loss: 0.1109 - accuracy: 0.9657 307/1875 [===>..........................] - ETA: 0s - loss: 0.1103 - accuracy: 0.9646 411/1875 [=====>........................] - ETA: 0s - loss: 0.1105 - accuracy: 0.9647 513/1875 [=======>......................] - ETA: 0s - loss: 0.1120 - accuracy: 0.9649 616/1875 [========>.....................] - ETA: 0s - loss: 0.1133 - accuracy: 0.9652 718/1875 [==========>...................] - ETA: 0s - loss: 0.1118 - accuracy: 0.9654 820/1875 [============>.................] - ETA: 0s - loss: 0.1129 - accuracy: 0.9651 923/1875 [=============>................] - ETA: 0s - loss: 0.1124 - accuracy: 0.96511027/1875 [===============>..............] - ETA: 0s - loss: 0.1105 - accuracy: 0.96591130/1875 [=================>............] - ETA: 0s - loss: 0.1102 - accuracy: 0.96601235/1875 [==================>...........] - ETA: 0s - loss: 0.1092 - accuracy: 0.96601343/1875 [====================>.........] - ETA: 0s - loss: 0.1084 - accuracy: 0.96621446/1875 [======================>.......] - ETA: 0s - loss: 0.1095 - accuracy: 0.96621549/1875 [=======================>......] - ETA: 0s - loss: 0.1094 - accuracy: 0.96621652/1875 [=========================>....] - ETA: 0s - loss: 0.1085 - accuracy: 0.96641755/1875 [===========================>..] - ETA: 0s - loss: 0.1082 - accuracy: 0.96651859/1875 [============================>.] - ETA: 0s - loss: 0.1077 - accuracy: 0.96671875/1875 [==============================] - 1s 487us/step - loss: 0.1076 - accuracy: 0.9666
<keras.src.callbacks.History at 0x2883a0e50>

6 evaluate

~accuracy: 0.9739 on testing data

Code 
model.evaluate(x_test,  y_test, verbose=2)
313/313 - 0s - loss: 0.0882 - accuracy: 0.9739 - 145ms/epoch - 464us/step
[0.08815735578536987, 0.9739000201225281]

7 predition

Code 
predict_x=model.predict(x_test) 
classes_x=np.argmax(predict_x,axis=1)
  1/313 [..............................] - ETA: 8s172/313 [===============>..............] - ETA: 0s313/313 [==============================] - 0s 288us/step

perdition first 10

Code 
classes_x[:10]
array([7, 2, 1, 0, 4, 1, 4, 9, 6, 9])

real first 10

Code 
y_test[:10]
array([7, 2, 1, 0, 4, 1, 4, 9, 5, 9], dtype=uint8)

8 save model

xxx.keras format not working at macos(2023-10-14) so using xxx.h5

Code 
model.save('my_model.h5',save_format='h5')

9 load model

Code 
new_model = tf.keras.models.load_model('my_model.h5')

# Show the model architecture
new_model.summary()
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 flatten (Flatten)           (None, 784)               0         
                                                                 
 dense (Dense)               (None, 128)               100480    
                                                                 
 dropout (Dropout)           (None, 128)               0         
                                                                 
 dense_1 (Dense)             (None, 10)                1290      
                                                                 
=================================================================
Total params: 101770 (397.54 KB)
Trainable params: 101770 (397.54 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

9.1 predition with load model

Code 
predict_x=new_model.predict(x_test) 
classes_x=np.argmax(predict_x,axis=1)
  1/313 [..............................] - ETA: 5s170/313 [===============>..............] - ETA: 0s313/313 [==============================] - 0s 291us/step

perdition first 10

Code 
classes_x[:10]
array([7, 2, 1, 0, 4, 1, 4, 9, 6, 9])

real first 10

Code 
y_test[:10]
array([7, 2, 1, 0, 4, 1, 4, 9, 5, 9], dtype=uint8)

10 Reference

https://www.tensorflow.org/tutorials/quickstart/beginner

https://en.wikipedia.org/wiki/MNIST_database