Let's use the most famous dataset here - the handwritten dataset - for model prediction.

In [1]:
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from collections import Counter
In [2]:
mnist = tf.keras.datasets.mnist
mnist
Out[2]:
<module 'keras.api.datasets.mnist' from 'E:\\7. Deep Learning\\venv\\lib\\site-packages\\keras\\api\\datasets\\mnist\\__init__.py'>
In [3]:
(train_image, train_label), (test_image, test_label) = mnist.load_data()
In [4]:
print('Shape of training images -  ', train_image.shape)
print('Label of training images -  ', train_label.shape)
print('Shape of testing images -  ', test_image.shape)
print('Shape of testing labels -  ', test_label.shape)

Shape of training images -   (60000, 28, 28)
Label of training images -   (60000,)
Shape of testing images -   (10000, 28, 28)
Shape of testing labels -   (10000,)
In [5]:
Counter(test_label)
Out[5]:
Counter({1: 1135,
         2: 1032,
         7: 1028,
         3: 1010,
         9: 1009,
         4: 982,
         0: 980,
         8: 974,
         6: 958,
         5: 892})
In [6]:
Counter(train_label)
Out[6]:
Counter({1: 6742,
         7: 6265,
         3: 6131,
         2: 5958,
         9: 5949,
         0: 5923,
         6: 5918,
         8: 5851,
         4: 5842,
         5: 5421})
In [7]:
np.set_printoptions(linewidth=200)
print(train_label[0])
print(train_image[0])

5
[[  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   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 [8]:
# Normalizing the images

train_image = train_image/255.0
test_image = test_image/255.0
In [18]:
# Building the model

model = tf.keras.models.Sequential([
    tf.keras.Input(shape=(28,28,1)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

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

model.summary()

Model: "sequential_1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ flatten_1 (Flatten)                  │ (None, 784)                 │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_3 (Dense)                      │ (None, 512)                 │         401,920 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_4 (Dense)                      │ (None, 128)                 │          65,664 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_5 (Dense)                      │ (None, 10)                  │           1,290 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 468,874 (1.79 MB)

 Trainable params: 468,874 (1.79 MB)

 Non-trainable params: 0 (0.00 B)
In [10]:
model.fit(train_image, train_label, epochs = 10)

Epoch 1/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 10s 5ms/step - accuracy: 0.9048 - loss: 0.3249
Epoch 2/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 8s 4ms/step - accuracy: 0.9762 - loss: 0.0787
Epoch 3/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 7s 4ms/step - accuracy: 0.9824 - loss: 0.0546
Epoch 4/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9882 - loss: 0.0386
Epoch 5/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9901 - loss: 0.0301
Epoch 6/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9917 - loss: 0.0242
Epoch 7/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 8s 4ms/step - accuracy: 0.9924 - loss: 0.0229
Epoch 8/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 7s 4ms/step - accuracy: 0.9943 - loss: 0.0164
Epoch 9/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 8s 4ms/step - accuracy: 0.9944 - loss: 0.0190
Epoch 10/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9949 - loss: 0.0155
Out[10]:
<keras.src.callbacks.history.History at 0x2161b86f790>
In [11]:
# Let's check the model summary after training

model.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ flatten (Flatten)                    │ (None, 784)                 │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense (Dense)                        │ (None, 512)                 │         401,920 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_1 (Dense)                      │ (None, 128)                 │          65,664 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_2 (Dense)                      │ (None, 10)                  │           1,290 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 1,406,624 (5.37 MB)

 Trainable params: 468,874 (1.79 MB)

 Non-trainable params: 0 (0.00 B)

 Optimizer params: 937,750 (3.58 MB)

What are all the other parameters and its values? Well I am giving a brief here. However, everything might not be clear as of now but it will surely be clear by the end of this course.
Optimizer Parameters
The "Optimizer params: 937,750" refers to the total number of parameters that the optimizer (e.g., Adam, SGD) maintains for updating the model. These include:

  1. Weights & Biases updates – The optimizer needs to store past updates for adaptive algorithms like Adam (momentum and adaptive learning rates).
  2. First & Second Moments (for Adam/RMSprop) – Adam keeps track of moving averages of gradients and squared gradients, which increases parameter count.
  3. Learning rate schedules (if used) – Some optimizers maintain additional parameters for learning rate adjustments.

Why Non-Trainable Params = 0?
Non-trainable parameters usually come from frozen layers (like pre-trained models where some layers are not updated). Since this model has zero non-trainable parameters, it means:

  • All layers are fully trainable (no frozen layers).
  • No batch normalization layers with fixed moving statistics.
  • No pre-trained weights used where some layers remain frozen.

Thus, all learnable parameters contribute to weight updates, and the optimizer manages additional memory for momentum-based updates.

Now, getting back to our model training. As you can see the accuracy has reached to 99 quiet before 10 epochs. So we dont need to go through all the 10 loops, and we can stop the training once the desired accuracy is reached using callbacks -

In [19]:
class mycallback(tf.keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs={}):
        if logs['accuracy'] >= 0.99:
            print('Reached 99% of accuracy. Stopping the iteration')
            self.model.stop_training = True
In [20]:
model.fit(train_image, train_label, epochs = 10, callbacks=[mycallback()])

Epoch 1/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9042 - loss: 0.3191
Epoch 2/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 6s 3ms/step - accuracy: 0.9737 - loss: 0.0818
Epoch 3/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 7s 4ms/step - accuracy: 0.9841 - loss: 0.0511
Epoch 4/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 7s 4ms/step - accuracy: 0.9881 - loss: 0.0370
Epoch 5/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9896 - loss: 0.0327
Epoch 6/10
1867/1875 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - accuracy: 0.9916 - loss: 0.0262Reached 99% of accuracy. Stopping the iteration
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 9s 5ms/step - accuracy: 0.9916 - loss: 0.0262
Out[20]:
<keras.src.callbacks.history.History at 0x2161b3c2da0>

So it stopped early

In [21]:
# Model evaluation
model.evaluate(test_image, test_label)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.9775 - loss: 0.0854
Out[21]:
[0.06842312961816788, 0.9818000197410583]
In [22]:
# Predictions
classified_images = model.predict(test_image)

313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 1ms/step
In [23]:
predictions = [np.argmax(images) for images in classified_images]
In [24]:
predictions[100], test_label[100]
Out[24]:
(6, 6)
In [25]:
predictions[1000], test_label[1000]
Out[25]:
(9, 9)

So what you did here?


  1. Used the Fashion Mnist dataset
  2. Looked into the basic data details
  3. Normalized the data
  4. You created the model architecture with -
    1. Input layer for the incoming data
    2. a Flatten layer
    3. 3 dense neural network layers with 512, 128 and 10 neurons alongwith activation functions
  5. Compiled the model with loss and optimizer
  6. Used a callback function in the fit and fitted the model with training data
  7. Evaluated the model with test data
  8. Predicted on the trained model
In [ ]: