How to build a simple neural network using tensorflow

Recipe Objective

How to build a simple neural network using tensorflow?

To build a simple neural network we require a dataset, and here we are going to use the “fashion-mnist” dataset which is already present in Keras. The dataset contains two sets which are the training set and test set, the training set is having 60,000 examples data where as the test is having 10,000 examples data. As we can also say we are having a data of 70,000 images of 28×28 size (28 pixels of width and 28 pixels of height) in which 60,000 is in training and 10,000 is in test. These images are in grayscale in which each one is showing the q0 possible clothing type.

Access Deep Learning Project to Build a Similar Images Finder in Python

Step 1 – Import library


import tensorflow as tf
from tensorflow import keras


Step 2 – Load Dataset


(x_train_data, y_train_data), (x_val_data, y_val_data) = keras.datasets.fashion_mnist.load_data()


Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
32768/29515 [=================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26427392/26421880 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
8192/5148 [===============================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4423680/4422102 [==============================] - 0s 0us/step

Here in the above we are loading the dataset, also we are splitting them into training set and validation set. x_train_data and y_train_data is here to build our model in which x_train_data will consist of image pixel data for 60,000 clothes whereas the y_train_data will consist of classes i.e clothing type for x_train_data. Similarly the x_val_data and y_val_data is used for testing or validation of our model.

Step 3 – Preprocess and create dataset


def preprocessing_function(x_new, y_new):
x_new = tf.cast(x_new, tf.float32) / 255.0
y_new = tf.cast(y_new, tf.int64)

def func_creating_dataset(xs_data, ys_data, num_classes=10):
ys_data = tf.one_hot(ys_data, depth=num_classes)
return tf.data.Dataset.from_tensor_slices((xs_data, ys_data)) \
.map(preprocessing_function) \


Step 4 – Create Neural network


dataset_training = func_creating_dataset(x_train_data, y_train_data)
dataset_val = func_creating_dataset(x_val_data, y_val_data)


Step 5 – Build the model


My_model = keras.Sequential([
keras.layers.Reshape(target_shape=(28 * 28,), input_shape=(28, 28)),
keras.layers.Dense(units=256, activation='relu'),
keras.layers.Dense(units=192, activation='relu'),
keras.layers.Dense(units=128, activation='relu'),
keras.layers.Dense(units=10, activation='softmax')
])


Step 6 – Train the model


My_model.compile(optimizer='adam',
loss=tf.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])

history = My_model.fit(
dataset_training.repeat(),
epochs=10,
steps_per_epoch=500,
validation_data=dataset_val.repeat(),
validation_steps=2


Epoch 1/10
500/500 [==============================] - 8s 12ms/step - loss: 0.6743 - accuracy: 0.7643 - val_loss: 0.4322 - val_accuracy: 0.8633
Epoch 2/10
500/500 [==============================] - 6s 11ms/step - loss: 0.3679 - accuracy: 0.8660 - val_loss: 0.3940 - val_accuracy: 0.8555
Epoch 3/10
500/500 [==============================] - 6s 12ms/step - loss: 0.3317 - accuracy: 0.8773 - val_loss: 0.3207 - val_accuracy: 0.9023
Epoch 4/10
500/500 [==============================] - 6s 12ms/step - loss: 0.3024 - accuracy: 0.8877 - val_loss: 0.3989 - val_accuracy: 0.8672
Epoch 5/10
500/500 [==============================] - 6s 12ms/step - loss: 0.2863 - accuracy: 0.8916 - val_loss: 0.3594 - val_accuracy: 0.8750
Epoch 6/10
500/500 [==============================] - 6s 12ms/step - loss: 0.2766 - accuracy: 0.8983 - val_loss: 0.4284 - val_accuracy: 0.8438
Epoch 7/10
500/500 [==============================] - 6s 11ms/step - loss: 0.2571 - accuracy: 0.9025 - val_loss: 0.2932 - val_accuracy: 0.8711
Epoch 8/10
500/500 [==============================] - 6s 12ms/step - loss: 0.2456 - accuracy: 0.9074 - val_loss: 0.1954 - val_accuracy: 0.9297
Epoch 9/10
500/500 [==============================] - 6s 12ms/step - loss: 0.2319 - accuracy: 0.9111 - val_loss: 0.3370 - val_accuracy: 0.8672
Epoch 10/10
500/500 [==============================] - 6s 11ms/step - loss: 0.2229 - accuracy: 0.9162 - val_loss: 0.3033 - val_accuracy: 0.9023

Step 7 – Make predictions


Make_predictions = My_model.predict(dataset_val)
Make_predictions


array([[8.21499402e-09, 1.98965409e-07, 1.17530963e-09, ...,
        2.39849047e-04, 3.75632886e-10, 9.99724329e-01],
       [4.75593419e-11, 1.32912683e-13, 1.02271674e-12, ...,
        3.79424998e-11, 1.00000000e+00, 1.16047255e-10],
       [6.36480451e-08, 1.22093002e-06, 1.70876291e-08, ...,
        2.29782687e-04, 4.84631757e-09, 9.99749601e-01],
       ...,
       [3.27635678e-13, 7.38353267e-11, 6.17022167e-14, ...,
        2.07368976e-06, 5.78794384e-15, 9.99997735e-01],
       [3.67982114e-15, 5.02441912e-18, 2.04865252e-15, ...,
        4.30711580e-18, 1.00000000e+00, 3.65123112e-20],
       [3.21948218e-10, 9.99999881e-01, 1.29861441e-10, ...,
        6.70993209e-12, 9.32399991e-11, 6.58160956e-11]], dtype=float32)

{“mode”:”full”,”isActive”:false}