natbusa / kernelgateway_demos

Some demo's to get you started exposing APIs from your jupyter notebooks.

Jupyter Gateway Demos

Some demo's to get you started exposing APIs from your jupyter notebooks.

Hello World

Check the basic demo on how to expose html, text, json data, and html images and snippets here:
https://github.com/natbusa/kernelgateway_demos/tree/master/hello-world

Iris classification

This demo shows how easy it is to explose train, and scoring ML functions straight from the notebook

The full API, including tuning, scoring, exploration exposing both json data as well as png images can be found on the following link: https://github.com/natbusa/kernelgateway_demos/blob/master/iris-classifier/notebooks/iris-classifier.ipynb

How to run

This demo requires docker (https://www.docker.com/community-edition#/download) to be installed in the system. Also use curl or postman or your web browser to inspect that the API exposed by the notebook is working correctly.

You need to build the container with the command:

docker build . -t iris_webapp_demo

Type the following to run the container and expose the webapp on port 9000:

docker run -t -p 9000:9000 iris_webapp_demo

Demo will run on http://localhost:9000/public/index.html

The full API, including tuning, scoring, exploration exposing both json data as well as png images can be found on the following link:
https://github.com/natbusa/kernelgateway_demos/blob/master/iris-classifier/notebooks/iris-classifier.ipynb

When finished with this demo, you can remove stop and remove the container by typing:

docker kill iris_webapp_demo
docker rm iris_webapp_demo

Notebook structure

In order to have the notebook working both in inline mode and in HTTP API serving mode with the kernel gateway we use an environment variable which is defined only while running the kernel gateway

KERNEL_GATEWAY = os.environ.get('KERNEL_GATEWAY', False)

Next to that we have number of utilities to export images as inline embedded html snippets. and we mock the request data structure so we can test inline in the notebook before running the gateway and exposing the api

The Jupyter Kernel Gateway also exposes the list of end points via a swagger interface on:
http://localhost:9000/_api/spec/swagger.json

For convenience the swagger UI (https://github.com/swagger-api/swagger-ui) has been added to this docker container: http://localhost:9000/public/swagger/index.html?url=http://localhost:9000/_api/spec/swagger.json

Endpoints

Explore Data: Scatter plot

GET /plots/scatter/:x/:y

where :x and :y are dimension index from 0 to 3 according to the following list:
['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)', 'petal width (cm)']

Try: http://localhost:9000/plots/scatter/0/1

Train Data

GET /model/train/:x/:y?model=:m

Will train the model according to two dimensions respectively :x, :y, and :m

where :x and :y are dimension index from 0 to 3 according to the following list:
['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)', 'petal width (cm)']

where :m from 0 to 5 fits one of the following models:

• KNN Classifier
• SVM classifier: linear kernel
• SVM classifier: gaussian kernel
• SVM classifier: polynomial kernel
• Decision Tree
• Decision Tree (depth limited)

For more information about this check
http://scikit-learn.org/stable/auto_examples/classification/plot_classifier_comparison.html

Try: http://localhost:9000/model/train/0/1?model=0,
should produce the following text, and train a knn model for the iris data

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
           metric_params=None, n_jobs=1, n_neighbors=3, p=2,
           weights='uniform')

Classify Data Points on the trained model

GET /model/predict/?variable=:v

Will classify the data according to the trained model,
where :v is a tuple of comma separated values, one for each dimension of the model

Try: http://localhost:9000/model/predict?variables=5.2,2.2 should score the tuple [5.2, 2.2] and produce the following text,

versicolor

Model precision, recall, and classify regions

GET /model/confusion/plot
GET /model/predict/plot

Try: http://localhost:9000/model/confusion/plot

Try: http://localhost:9000/model/predict/plot

Web App

You can use your favourite js and css framework to render the API exposed by the Jupyter Kernel Gateway. This demo uses a minimal jquery framework with backgroung ajax calls and where data, once received is injected in the html DOM.

Have a look at the js:
https://github.com/natbusa/kernelgateway_demos/blob/master/iris-classifier/static/assets/js/script.js

Here is the jquery code which trains the model and generate the confusion matrix and the classification overlay pic.

$("button#train").click(function(){

    var var_x = $("#var_x").val();
    var var_y = $("#var_y").val();
    var model = $("#model").val()
    var path = "/model/train/"+var_x+"/"+var_y+"?model="+model

    $.get(path).done(function(data, status){
        $("#train_output").html("Model: " + data);

        var p1 = $.get("/model/confusion/plot");
        var p2 = $.get("/model/predict/plot");

        $.when(p1, p2).done(function(r1, r2) {
            $("#predict_plot").html(r1[0]);
            $("#confusion_plot").html(r2[0]);

            $("#predict_form").show()
        });

    });
});

Resources

https://github.com/jupyter/kernel_gateway_demos
https://github.com/jupyter/kernel_gateway/tree/master/etc/api_examples