to get the requirements.txt file in conda env:

pip list --format=freeze > requirements.txt

---

Prerequisite:

• Install and run Docker Desktop on Windows before executing 'docker' commands

Docker Containers:

Training Container:

Build the training container

docker build -t transfer_train-container -f Dockerfile_train .

check gpus on training container

docker run --gpus 1 -ti transfer_train-container nvidia-smi

Run the training container

• in unix terminal:

docker run --gpus 1 -v $(pwd)/data:/home/prgrmcode/app/data -ti --name train-container transfer_train-container

• in windows command prompt:

docker run --gpus 1 -v "%cd%/data:/home/prgrmcode/app/data" -ti --name train-container transfer_train-container

-- number of gpus, -v --volume mounts first folder from local machine to the folder in docker container, -ti target image, command(python3 .py)

Application Container:

Build the application container

docker build -t transfer_app-container -f Dockerfile_app .

Run the application container

docker run -it --gpus 1 -p 5000:5000 --name app-container transfer_app-container bash

---

---

With Docker compose:

Run everything easily from docker-compose.yml file with one command

To create and run new training and app container together:

docker-compose up

When train and app container are up and running, you can navigate to:

• localhost:5000

• app predictions

Create and run a new training container:

docker-compose up --no-deps --build transfer_train-container

Create and run a new app container for hosting the application:

docker-compose up --build transfer_app-container

To start / reuse the existing training container:

docker-compose start transfer_train-container

To start / reuse the existing app container:

docker-compose start transfer_app-container

If docker uses too much disk space, run:

docker system prune

---