Here I show how to use tensorflow keras library to design an AI model that can identify handwritten digits
Reading dataset¶
Lets read MNIST dataset and display first few images
In [1]:
import tensorflow as tf
from tensorflow.keras.datasets import mnist
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
WARNING:tensorflow:From C:\Users\hi\Desktop\projects\python_projects\tutorial\tut_tensorflow\.venv\Lib\site-packages\keras\src\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead.
Step 1: Load the MNIST dataset.¶
It has 60,000 images for training and 10,000 images for testing.
In [2]:
(X_train, y_train), (X_test, y_test) = mnist.load_data()
# Step 2: Display the shape of the dataset
print(f"X_train shape: {X_train.shape}") # (60000, 28, 28).
print(f"y_train shape: {y_train.shape}") # (60000,)
print(f"X_test shape: {X_test.shape}") # (10000, 28, 28)
print(f"y_test shape: {y_test.shape}") # (10000,)
X_train shape: (60000, 28, 28) y_train shape: (60000,) X_test shape: (10000, 28, 28) y_test shape: (10000,)
In [ ]:
EDA¶
Lets try to see and visualize our data.
In [3]:
# Lets look at the first image data and also plot it. We are trybing to understand the image data
img_idx=0
print(X_train[img_idx].shape)
print("--------------------")
np.set_printoptions(linewidth=150, precision=2, suppress=True) # Adjust formatting
print(X_train[img_idx])
print("*********************")
plt.figure(figsize=(8,8)) # Set the figure size
plt.imshow(X_train[img_idx],
cmap='gray' # # Display the image in grayscale. comment it out if u want to see color
)
plt.title(f"Label: {y_train[img_idx]}") # Set the title as the label
plt.axis('off') # Turn off axis
plt.tight_layout()
plt.show()
(28, 28) -------------------- [[ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 3 18 18 18 126 136 175 26 166 255 247 127 0 0 0 0] [ 0 0 0 0 0 0 0 0 30 36 94 154 170 253 253 253 253 253 225 172 253 242 195 64 0 0 0 0] [ 0 0 0 0 0 0 0 49 238 253 253 253 253 253 253 253 253 251 93 82 82 56 39 0 0 0 0 0] [ 0 0 0 0 0 0 0 18 219 253 253 253 253 253 198 182 247 241 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 80 156 107 253 253 205 11 0 43 154 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 14 1 154 253 90 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 139 253 190 2 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 11 190 253 70 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 35 241 225 160 108 1 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 81 240 253 253 119 25 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 45 186 253 253 150 27 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 16 93 252 253 187 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 249 253 249 64 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 46 130 183 253 253 207 2 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 39 148 229 253 253 253 250 182 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 24 114 221 253 253 253 253 201 78 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 23 66 213 253 253 253 253 198 81 2 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 18 171 219 253 253 253 253 195 80 9 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 55 172 226 253 253 253 253 244 133 11 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 136 253 253 253 212 135 132 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]] *********************
In [6]:
# Lets visualize above using heatmap. # TODO Fix the values that are displayed in scientific notation
plt.figure(figsize=(12, 12))
sns.heatmap(X_train[img_idx], annot=True, fmt="03d", cmap="gray", cbar=False)
plt.show()
Lets understand above data. The first image, which is 5, has 28 rows and 28 columns. It is so because this is image size 28 X 28. Each value in the row,col is from 0-255, which is color range. 0 being black and 255 being white
In [ ]:
In [5]:
# Display the first 6 images with their labels
plt.figure(figsize=(12, 6)) # Set the figure size
for i in range(6):
plt.subplot(2, 3, i + 1) # Create a 2x3 grid for 6 images
plt.imshow(X_train[i], cmap='gray') # Display the image in grayscale
plt.title(f"Label: {y_train[i]}") # Set the title as the label
plt.axis('off') # Turn off axis
plt.tight_layout()
plt.show()
In [7]:
# Now we convert the 28X28 image data, which is a 2D array, into 1D array whose size would be 28*28=784.
# We do this becuase the input layer of Neural Network accepts only 1D data.
X_train = X_train.reshape(-1, 784).astype('float32') / 255.0 # Flatten and normalize training data
X_test = X_test.reshape(-1, 784).astype('float32') / 255.0 # Flatten and normalize test data
print(X_train.shape)
print(X_test.shape)
(60000, 784) (10000, 784)
In [9]:
#Lets look at the value of 1st image now and compare with above
print(X_train[0]) # This is 1D array whose size is 784
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.01 0.07 0.07 0.07 0.49 0.53 0.69 0.1 0.65 1. 0.97 0.5 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.12 0.14 0.37 0.6 0.67 0.99 0.99 0.99 0.99 0.99 0.88 0.67 0.99 0.95 0.76 0.25 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.19 0.93 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.98 0.36 0.32 0.32 0.22 0.15 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.07 0.86 0.99 0.99 0.99 0.99 0.99 0.78 0.71 0.97 0.95 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.31 0.61 0.42 0.99 0.99 0.8 0.04 0. 0.17 0.6 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.05 0. 0.6 0.99 0.35 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.55 0.99 0.75 0.01 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.04 0.75 0.99 0.27 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.14 0.95 0.88 0.63 0.42 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.32 0.94 0.99 0.99 0.47 0.1 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.18 0.73 0.99 0.99 0.59 0.11 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.06 0.36 0.99 0.99 0.73 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.98 0.99 0.98 0.25 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.18 0.51 0.72 0.99 0.99 0.81 0.01 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.15 0.58 0.9 0.99 0.99 0.99 0.98 0.71 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.09 0.45 0.87 0.99 0.99 0.99 0.99 0.79 0.31 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.09 0.26 0.84 0.99 0.99 0.99 0.99 0.78 0.32 0.01 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.07 0.67 0.86 0.99 0.99 0.99 0.99 0.76 0.31 0.04 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.22 0.67 0.89 0.99 0.99 0.99 0.99 0.96 0.52 0.04 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.53 0.99 0.99 0.99 0.83 0.53 0.52 0.06 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. ]
In [ ]:
In [ ]:
Creating a Simple Neural Network using tf.keras¶
tf.keras is TensorFlow’s high-level API to build and train deep learning models. You can define models using the Sequential class for a simple stack of layers
In [7]:
# Lets create a simple model which takes input 28*28=784 features, and produces output that are in 10 classes(0-9)
import tensorflow as tf
from tensorflow.keras import layers, models
# Create a simple sequential model
# Input layer (28*28=784 features). 64 is the number of neuron in hidden fully connected layer (dense layer)
model = models.Sequential([
layers.Dense(64, activation='relu', input_shape=(784,)),
layers.Dense(10, activation='softmax') # Output layer (10 classes)
])
c:\Users\hi\Desktop\projects\python_projects\tutorial\tut_tensorflow\.venv\Lib\site-packages\keras\src\layers\core\dense.py:106: 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)
In [8]:
# Lets visualize
from tensorflow.keras.utils import plot_model
plot_model(
model,
to_file='model.png',
show_shapes=True,
show_layer_names=True,
rankdir='LR' # TB Top to Bottom (vertical)
)
Out[8]:
In [9]:
# Compile the model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
# Model summary to see the architecture
model.summary()
Model: "sequential"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩ │ dense (Dense) │ (None, 64) │ 50,240 │ ├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤ │ dense_1 (Dense) │ (None, 10) │ 650 │ └──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 50,890 (198.79 KB)
Trainable params: 50,890 (198.79 KB)
Non-trainable params: 0 (0.00 B)
In [ ]:
Train above model¶
Lets use above model
To train a model, you need data. You can use tf.data to load datasets efficiently, and you can train models using the .fit() method on the dataset.
Example (Training a model on MNIST data): ( took 2 minutes)
In [9]:
# 60000/32 = 1688
# Train the model
model.fit(X_train, y_train, epochs=5, batch_size=32, validation_split=0.1) # Train for 5 epochs. Took 30 secs
# Evaluate the model
test_loss, test_acc = model.evaluate(X_test, y_test)
print(f"Test accuracy: {test_acc:.2f}")
Epoch 1/5 1688/1688 ━━━━━━━━━━━━━━━━━━━━ 6s 3ms/step - accuracy: 0.8456 - loss: 0.5361 - val_accuracy: 0.9553 - val_loss: 0.1613 Epoch 2/5 1688/1688 ━━━━━━━━━━━━━━━━━━━━ 4s 3ms/step - accuracy: 0.9497 - loss: 0.1733 - val_accuracy: 0.9655 - val_loss: 0.1213 Epoch 3/5 1688/1688 ━━━━━━━━━━━━━━━━━━━━ 4s 3ms/step - accuracy: 0.9656 - loss: 0.1173 - val_accuracy: 0.9728 - val_loss: 0.0940 Epoch 4/5 1688/1688 ━━━━━━━━━━━━━━━━━━━━ 4s 3ms/step - accuracy: 0.9747 - loss: 0.0873 - val_accuracy: 0.9715 - val_loss: 0.0982 Epoch 5/5 1688/1688 ━━━━━━━━━━━━━━━━━━━━ 4s 3ms/step - accuracy: 0.9783 - loss: 0.0711 - val_accuracy: 0.9730 - val_loss: 0.0928 313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.9656 - loss: 0.1080 Test accuracy: 0.97
In [ ]:
# Make predictions on one image
predictions = model.predict(X_test[0].reshape(1, -1)) # Predict on the first sample
predicted_labels = predictions.argmax(axis=1)
print("Predictions for the first test sample:")
print("Predicted label:", predicted_labels[0]) # Print predicted class label
print("True label:", y_test[0]) # Print true class label
# Reshape and display the first image
x_test_image = X_test[0].reshape(28, 28) # Reshape back to 28x28 for visualization
plt.imshow(x_test_image, cmap='gray') # Display the image
plt.title(f"Pred: {predicted_labels[0]}\nTrue: {y_test[0]}")
plt.axis('off')
plt.tight_layout()
plt.show()
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step Predictions for the first test sample: Predicted label: 7 True label: 7
In [ ]:
# Make predictions on 1st 5 test images
predictions = model.predict(X_test[:5]) # Predict on the first 5 samples of the test set
predicted_labels = predictions.argmax(axis=1)
print("Predictions for the first 5 test samples:")
print(predicted_labels) # Print predicted class labels
print("True labels:", y_test[:5]) # Print true class labels
# Display the first 5 images with predictions
x_test_images = X_test[:5].reshape(-1, 28, 28) # Reshape back to 28x28 for visualization
plt.figure(figsize=(10, 5))
for i in range(5):
plt.subplot(1, 5, i + 1) # Create subplots for 5 images
plt.imshow(x_test_images[i], cmap='gray') # Display the image
plt.title(f"Pred: {predicted_labels[i]}\nTrue: {y_test[i]}")
plt.axis('off')
plt.tight_layout()
plt.show()
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 182ms/step Predictions for the first 5 test samples: [7 2 1 0 4] True labels: [7 2 1 0 4]
In [ ]:
Create and train different model on the same dataset¶
In [12]:
from tensorflow.keras import layers, models
# Load and prepare the MNIST dataset
(X_train, y_train), (X_test, y_test) = mnist.load_data()
# Normalize the data
X_train = X_train / 255.0
X_test = X_test / 255.0
# X_train, X_test = X_train / 255.0, X_test / 255.0
# Build a simple model (as we did previously)
model = models.Sequential([
layers.Flatten(input_shape=(28, 28)), # Flatten 28x28 images into 1D
layers.Dense(128, activation='relu'), # create a dense layer of 128 neuron
layers.Dropout(0.2), # randomly drop 20% of neuron to prevent overfitting
layers.Dense(10, activation='softmax')
])
# Compile the model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
# Train the model on the MNIST training data
model.fit(X_train, y_train, epochs=5)
Epoch 1/5 1875/1875 ━━━━━━━━━━━━━━━━━━━━ 6s 3ms/step - accuracy: 0.8601 - loss: 0.4780 Epoch 2/5 1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 3ms/step - accuracy: 0.9557 - loss: 0.1490 Epoch 3/5 1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 3ms/step - accuracy: 0.9674 - loss: 0.1057 Epoch 4/5 1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 3ms/step - accuracy: 0.9722 - loss: 0.0898 Epoch 5/5 1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 3ms/step - accuracy: 0.9775 - loss: 0.0715
Out[12]:
<keras.src.callbacks.history.History at 0x25cb6c54490>
In [16]:
# Evaluate the model on test data
test_loss, test_acc = model.evaluate(X_test, y_test)
print(f'Test accuracy: {test_acc}')
313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.9698 - loss: 0.1009 Test accuracy: 0.9758999943733215
By increasing model hidden layers and adding drop out layer has increased the accuracy little bit.
In [ ]:
Saving and Loading Models¶
Once a model is trained, you can save it to disk and load it later for inference or retraining.
Example:
In [ ]:
# Save the model
model.save("model.keras")
In [ ]:
# Load the model back
loaded_model = tf.keras.models.load_model("model.keras")
# Evaluate the loaded model
test_loss, test_acc = loaded_model.evaluate(X_test, y_test)
print(f'Loaded model test accuracy: {test_acc}')
313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 984us/step - accuracy: 0.9722 - loss: 0.0893 Loaded model test accuracy: 0.9767000079154968
In [ ]:
In [ ]:
In [ ]: