Classification of Numbers(0-9)

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np

(X_train, y_train) , (X_test, y_test) = keras.datasets.mnist.load_data()

Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11490434/11490434 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step

X_train.shape, y_train.shape

((60000, 28, 28), (60000,))

X_test.shape, y_test.shape

((10000, 28, 28), (10000,))

plt.imshow(X_train[5])
y_train[5]

2

X_train = X_train / 255 # In order to normalize the data
X_test = X_test / 255

X_train_flat = X_train.reshape(len(X_train), 28*28)
X_test_flat = X_test.reshape(len(X_test), 28*28)

X_train_flat.shape, X_test_flat.shape

((60000, 784), (10000, 784))

model = keras.Sequential([
    keras.layers.Dense(10, input_shape=(784,), activation='sigmoid')
])

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(X_train_flat, y_train, epochs=5)

/usr/local/lib/python3.10/dist-packages/keras/src/layers/core/dense.py:87: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(activity_regularizer=activity_regularizer, **kwargs)

Epoch 1/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 2ms/step - accuracy: 0.8194 - loss: 0.7002
Epoch 2/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 4s 2ms/step - accuracy: 0.9164 - loss: 0.3028
Epoch 3/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 4s 2ms/step - accuracy: 0.9198 - loss: 0.2839
Epoch 4/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 1ms/step - accuracy: 0.9235 - loss: 0.2728
Epoch 5/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9255 - loss: 0.2593

<keras.src.callbacks.history.History at 0x7f40f2dd82b0>

Given the input dataset, the accuracy is around 93% with a very simple neural network model. The model complex is less yet achieved good accuracy.

model.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ dense (Dense)                        │ (None, 10)                  │           7,850 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 23,552 (92.00 KB)

 Trainable params: 7,850 (30.66 KB)

 Non-trainable params: 0 (0.00 B)

 Optimizer params: 15,702 (61.34 KB)

model.evaluate(X_test_flat, y_test)

313/313 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.9144 - loss: 0.3048

[0.27070537209510803, 0.9248999953269958]

y_predict = model.predict(X_test_flat)
y_predict[0]

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step

array([4.6898533e-02, 2.9230904e-07, 6.8001457e-02, 9.5464110e-01,
       2.4680649e-03, 1.2394587e-01, 1.4552444e-06, 9.9982125e-01,
       8.0362149e-02, 5.2514255e-01], dtype=float32)

plt.imshow(X_test[99])

np.argmax(y_predict[99])

9

#Let's make our mode more complex by adding a hidden layer

model = keras.Sequential([
    keras.layers.Dense(100, input_shape=(784,), activation='relu'),
    keras.layers.Dense(10, activation='sigmoid')
])

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(X_train_flat, y_train, epochs=5)

Epoch 1/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 6s 2ms/step - accuracy: 0.8660 - loss: 0.4641
Epoch 2/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 2ms/step - accuracy: 0.9597 - loss: 0.1381
Epoch 3/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 4s 2ms/step - accuracy: 0.9740 - loss: 0.0886
Epoch 4/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9794 - loss: 0.0687
Epoch 5/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9842 - loss: 0.0515

<keras.src.callbacks.history.History at 0x7f40f2d75e10>

model.summary()

Model: "sequential_1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ dense_1 (Dense)                      │ (None, 100)                 │          78,500 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_2 (Dense)                      │ (None, 10)                  │           1,010 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 238,532 (931.77 KB)

 Trainable params: 79,510 (310.59 KB)

 Non-trainable params: 0 (0.00 B)

 Optimizer params: 159,022 (621.18 KB)

Well there is a lot of improvement after adding a layer. This is a good model.

model.evaluate(X_test_flat,y_test)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.9711 - loss: 0.0966

[0.08135204017162323, 0.9746999740600586]

# Let's look at some results.

plt.imshow(X_test[9])
np.argmax(y_predict[9])

9

plt.imshow(X_test[15])
np.argmax(y_predict[15])

5

plt.imshow(X_test[987])
np.argmax(y_predict[987])

7