https://github.com/marcelo-earth/fashionitem-classifier

👚⚙️ A fashion-item classifier developed with TensorFlow 2.0.0
https://github.com/marcelo-earth/fashionitem-classifier

keras neural-networks python tensorflow2

Last synced: 9 days ago
JSON representation

👚⚙️ A fashion-item classifier developed with TensorFlow 2.0.0

• Host: GitHub
• URL: https://github.com/marcelo-earth/fashionitem-classifier
• Owner: marcelo-earth
• Created: 2019-11-24T21:24:28.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2021-10-23T04:55:33.000Z (about 3 years ago)
• Last Synced: 2024-10-21T01:52:58.012Z (18 days ago)
• Topics: keras, neural-networks, python, tensorflow2
• Language: Jupyter Notebook
• Homepage:
• Size: 80.1 KB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
👚 Fashion Item Classifier ⚙️


A fashion item classifier developed with TensorFlow 2.0.0 and Keras.




  

  

  

  

  

  



It is based of the popular [`fashion_mnist`](https://github.com/zalandoresearch/fashion-mnist) dataset of Keras:

![MNIST Dataset](https://github.com/zalandoresearch/fashion-mnist/raw/master/doc/img/fashion-mnist-sprite.png)

![TensorFlow](./.github/tensorflow.png)

## 📦 Deployment

It's important install **TensorFlow 2.0.0** with PIP using:

```

pip install tensorflow==2.0.0-alpha0

```

Then install other libraries like:

* Pandas

* Numpy

* Matplotlib

With its latest version using PIP.

## 🚀 How it works?

### 📊 Reading data

Load MNIST Fashion-item dataset:

```python

data = keras.datasets.fashion_mnist

(train_images, train_labels), (test_images, test_labels) = data.load_data()

```

The dataset split `train_images` and `test_images` are numpy type array with 784 numbers from 0 to 255. These arrays can be converted into images of 28 by 28 pixels (that's why 784 numbers).

Then reduce the data dividing by 225, beacuse is a good practice shrink our data. Now the values of 225 is equal to 1.0 and so on:

```python

train_images = train_images / 255.0

test_images = test_images / 255.0

```

Based on the [labels](https://github.com/zalandoresearch/fashion-mnist#labels) for each fashion item, create a list for ten elements:

```python

class_names = ['T-shirt/top',

               'Trouser',

               'Pullover',

               'Dress',

               'Coat',

               'Sandal',

               'Shirt',

               'Sneaker',

               'Bag',

               'Ankle boot']

```

### 🧠 Neural Network Development

Using `Sequential` class from Keras and add layers.

1. The first layer is flattening input data based on 28x28 pixels of the image.

2. Then, the second layer receive the data from the previous layer and works with it with 128 neurons using RELU activation.

3. The final layer receive the data from the previous layer and works with an ouput of 10 (based on the 10 fashion-item) using SoftMax activation.

```python

model = keras.Sequential([

  keras.layers.Flatten(input_shape=(28, 28)),

  keras.layers.Dense(128, activation="relu"),

  keras.layers.Dense(10, activation="softmax")

])

```

### Building the Neural Network

Using `compile` method of `Sequential` using the necessary parameters: 

* **Optimizer:** Set as [`adam`](https://towardsdatascience.com/adam-latest-trends-in-deep-learning-optimization-6be9a291375c), an optimizing algorithm.

* **Loss:** Set as `sparse_categorical_crossentropy`

```python

model.compile(

  optimizer="adam",

  loss="sparse_categorical_crossentropy",

  metrics=["accuracy"]

)

```