Fraud-Detection-Production

It is basically extension to my previous work where I used tensorflow pipelines (etl) to enhance the efficiency of cpu-gpu parallel processing

This repository contains prod-code where we used autoencoder to detect frauds in transaction made on Tensorflow v1.13.1 framework with tensorflow serving api v1.13.1

structure for the files and code :

checkpoints -> model checkpoints while training

data -> both processed and raw

deployment -> tensorflow serving api

model-export -> saved model

notebook-> Data exploration and working protoype (training the model with tf pipeline)

src-> source code

summary-> model summary while training and validation on test set

dependcies : tensorflow v1.13.1

how to run the api

1> old school way

install tensorflow server model and port to 8500 for grpc request and give model_name and model_path and install

tensorflow_model_server --port=8500 --model_name=anamoly_detection --model_base_path=$HOME/Desktop/Fraud-Detection-Production-master/model-export/anamoly_detection/

tensorflow serving api v1.13.1 with that

run -> python client.py

2nd way docker containerization approach though I deployed whole model in aws but cost way up high you can use this way which very easy

1> download docker image for tensorflow serving api of google which has all dependecies

2> create it's container

docker create -p 8500:8500 -e MODEL_NAME=anamoly_detection --mount type=bind , source=$HOME/Desktop/Fraud-Detection-Producton/model-export/anamoly_detection,target=/models/anamoly_detection --name=my_container1 tensorflow/serving

3> start the container once container is started you don't need to run sever again and again

docker start my_container1

4> run python client.py