access-niu

This repository contains application to train models for Image classification and Regression tasks.

Tasks

[x] Create a basic app for training and inference
[ ] Support for training regression models
[ ] Support of training multi input/output models
[ ] Incorporate Bayesian Inference for finding uncertainty in the predictions.
[x] Create Docker Image
[ ] Support for serving the application with gunicorn
[ ] Anything else?

Installation

pip install access-niu

Training

python -m access_niu.train --template sample/colors/template.yml

Inference

python -m access_niu --projects output

Now use this curl command to parse

curl -X POST \
  http://localhost:8000/parse \
  -F data=@samples/colors/train/red/1.jpg

Docker

Build image

clone the git repo

git clone https://github.com/accessai/access-niu.git

Build the docker image
```
docker build -t access-niu:latest .
```
Run the docker container.

Note: You can attach a directory as a volume so that you can supply the templates from outside the docker container.


# we will use it as root directory for access-niu application
mkdir accessai
# copy samples folder
cp -r samples accessai/

train the model

docker run -v $(pwd)/accessai:/accessai access-niu python -m access_niu.train --template samples/colors/template.yml

After running the train command you should get an output folder in the accessai directory

Now start the access_niu server
```bash
docker -d run -v $(pwd)/accessai:/accessai -p 8000:8000 access-niu --projects output