Please star this repo if you are using it as a reference. Thanks.

This is a PoC for creating a pipeline that does the following -

• Training a model. Here it is done on Fashion MNIST and MNIST datasets.
• Load the saved weights.
• Expose an API using Flask.

• OS - Ubuntu 18.04 LTS
• Python Version - 3.6
• Conda Version - Miniconda3-latest-Linux-x86_64
• Shell Flavour - ZSH (Optional)
• Tensorflow Version - 1.13.1
• Keras Version - keras-2.2.4

• Install Miniconda from here.
• Run the following command

conda update conda

• Clone the repository
• Install dependencies from environment.yml file

conda env create -f environment.yml

• Activate your environment

conda activate classification_model

Refer the doc here for knowing what all flavours are supported.

• Visit src/train folder and run the following command

python3 fashion_mnist.py

You will get a h5 model and a json file as an output in the model folder. You can do the same thing for MNIST dataset.

• Now go to src/server and start the server using the following command

python3 server.py

For MNIST dataset go to pre-process folder and change json_file and load_weights accordingly

• Open a new temrinal and visit src/test for images and test according to the dataset you loaded by using the following command

curl -X POST -F '[email protected]'  http://localhost:5000/predict

For other images, in place of trouser.jpg add the name of that new input image.

Numbers are returned in both the cases. In case of MNIST dataset, it directly corresponds to the predicted number whereas in case of Fashion MNIST dataset, it corresponds to the class number. For more information refer this link.

NOTE : Since python2 is being deprecated and is going to go out of service on January 1st, 2020, hence the code was made for python3. Depending on the default version of python in your system you may or may not have to run the command as python3 fashion_mnist.py. It can be python fashion_mnist.py given that default version is python3.