How it works

Prerequisites

• Docker is up and running
• Git installed

Deploy an environment for a single use

git clone --recursive https://github.com/4-DS/sinara-ext-tools.git
cd sinara-ext-tools

To make use of it, run:

bash create.sh
bash run.sh

Go on http://127.0.0.1:8888/lab

git clone --recursive https://github.com/4-DS/step_template.git
cd step_template

Run 'Init_Data.ipynb' to get sample data

Run 'step.dev.py' in Terminal

python step.dev.py

To stop using it for a while, run:

bash stop.sh

To continue using it, run:

bash run.sh

To remove it, run:

bash remove.sh

Let's create a simple ML pipeline

Once your Sinara single use was deployed, you should create Git repositories for your ML pipeline's steps. Each step is based on this template repository https://dev.azure.com/swat-team/mlbox/_git/mlbox_step_template by using README.md In each step you must define:

• inputs
• outputs
• custom_inputs
• custom_outputs
• tmp_inputs
• tmp_outputs

Inputs are some previous steps outputs. Outputs are some results of a step. Inputs/outputs are formed base on a special run name which is 'run-%timestamp%'

Custom inputs/outputs

See the ready steps step1-4 of pipeline with the name 'pipeline0' at

1. https://github.com/4-DS/pipeline-step1.git
2. https://github.com/4-DS/pipeline-step2.git
3. https://github.com/4-DS/pipeline-step3.git
4. https://github.com/4-DS/pipeline-step4.git

Then you can see design of your ML pipeline by running visualize.ipynb

Download it in the root folder, containing all your steps, set parameters and run

Let's build production image with your model

Please, download the ready ML model example:

1. https://github.com/4-DS/pipeline-model_train.git

Run python step.dev.py Then pick up the entity path for your model packed as a bentoservice entity Then run bash containerize.sh and set parameters

Now you get an image with your model ready for intergration with your environment