Handwritten Digits - Python Notebook

Handwritten digits are a classic dataset in machine learning and computer vision. They consist of images of handwritten digits (0-9) and are commonly used for training and testing image classification algorithms.

Importing Libraries and Dataset¶

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
print(x_train.shape)

(60000, 28, 28)

plt.matshow(x_train[0])
print(y_train[0])

x_train = x_train / 255.0
x_test = x_test / 255.0

Model¶

model = keras.Sequential(
    [
        keras.layers.Flatten(input_shape=(28, 28)),
        keras.layers.Dense(100, activation="relu"),
        keras.layers.Dense(10, activation="sigmoid"),
    ]
)

model.summary()

/usr/lib/python3.13/site-packages/keras/src/layers/reshaping/flatten.py:37: 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__(**kwargs)

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

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

Epoch 1/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9243 - loss: 0.2686
Epoch 2/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9643 - loss: 0.1224
Epoch 3/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9750 - loss: 0.0849
Epoch 4/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9808 - loss: 0.0638
Epoch 5/5
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9843 - loss: 0.0505

<keras.src.callbacks.history.History at 0x7ff332dba520>

Evaluation¶

model.evaluate(x_test, y_test)

313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 935us/step - accuracy: 0.9753 - loss: 0.0798

[0.0798436775803566, 0.9753000140190125]

y_pred_labels = [np.argmax(pred) for pred in model.predict(x_test)]

313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 577us/step

cm = tf.math.confusion_matrix(labels=y_test, predictions=y_pred_labels)
cm

<tf.Tensor: shape=(10, 10), dtype=int32, numpy=
array([[ 959,    0,    1,    2,    3,    6,    2,    1,    3,    3],
       [   0, 1127,    3,    1,    0,    1,    1,    0,    2,    0],
       [   3,    4, 1007,    2,    2,    0,    1,    4,    9,    0],
       [   0,    1,    4,  991,    1,    2,    0,    5,    2,    4],
       [   0,    0,    3,    0,  971,    0,    2,    0,    1,    5],
       [   1,    1,    0,    9,    1,  874,    3,    0,    1,    2],
       [   2,    3,    1,    2,   13,   11,  921,    0,    5,    0],
       [   0,    5,   11,    2,    4,    1,    1,  990,    4,   10],
       [   1,    1,    3,    7,    4,    9,    1,    3,  943,    2],
       [   1,    6,    0,    7,   14,    5,    1,    2,    3,  970]],
      dtype=int32)>

plt.figure(figsize=(10, 7))
sns.heatmap(cm, annot=True, fmt="d")
plt.xlabel("Predicted")
plt.ylabel("Truth")