fastai-text-classifier-server

A simple Flask server for a fast.ai text classifier model. Included is the SMS classifier notebook which I used to train the model. I'll be adding details to the deploy steps.

Example:

0.0.0.0:8080/predict?sms=this%20is%20a%20text

{
  "accuracy": 0.9927310347557068, 
  "prediction": "good"
}

Setup

Train the model

1. Setup a jupyter notebook project with the sms-classifier.ipynb notebook.
2. Change the hushed directory name to something that suits your application.
3. Download the Kaggle sms spam dataset: https://www.kaggle.com/uciml/sms-spam-collection-dataset
4. (Optional) Add your own SMS collections and labels.
5. Train your text classifier, at the end of the notebook the model will be exported as export.pkl.
6. Copy the export.pkl file into the data directory of the flask app.

Deploy wih AWS ECS

Local build

You'll need to have docker installed on your machine to get through this part.

1. Build the docker image: docker build -t TextClassifier:latest
2. Run the docker image and make sure it works: docker run -d -p 80:80 TextClassifier:latest, it'll be running on 0.0.0.0.

Upload to ECR

You'll need an AWS account to get through this part.

1. Search for ECR on the AWS web console, and then create a repository which will hold your docker image.
2. Click on view push commands and follow the instructions to upload your image.
3. Follow Amazon's directions for deploying a docker container: https://aws.amazon.com/getting-started/tutorials/deploy-docker-containers/ . When you create the task definition paste in the image URI of the docker image you uploaded to ECR.

End Result

https://www.quietsheriff.com/