Devskiller programming task sample - Machine Learning

Introduction

With Devskiller.com you can assess your candidates' programming skills as a part of your recruitment process. We have found that programming tasks are the best way to do this and have built our tests accordingly. The way our test works is your candidate is asked to modify the source code of an existing project.

During the test, your candidates have the option of using our browser-based code editor and can build the project inside the browser at any time. If they would prefer to use an IDE they are more comfortable with, they can also download the project code or clone the project’s Git repository and work locally.

You can check out this short video to see the test from the candidate's perspective.

This repo contains a sample project for Python and below you can find a detailed guide for creating your own programming project.

Please make sure to read our Getting started with programming projects guide first

Technical details

Any project which uses pip and setuptools can be used as a programming task. We use pytest to execute unit tests.

Your project will be executed with following command:

pip install -e . && python3 setup.py pytest

Check out our setup.py and setup.cfg files.

Automatic assessment

It is possible to automatically assess the solution posted by the candidate. Automatic assessment is based on unit tests results and code quality measurements.

There are two kinds of unit tests:

1. Candidate tests - unit tests that the candidate can see during the test should be used only for basic verification and to guide the candidate in understanding the requirements of the project. Candidate tests WILL NOT be used to calculate the final score.
2. Verification tests - unit tests that the candidate can’t see during the test. Files containing verification tests will be added to the project after the candidate finishes the test and will be executed during the verification phase. The results of the verification tests will be used to calculate the final score.

Once the solution is developed and submitted, the platform executes verification tests and performs static code analysis.

Devskiller project descriptor

Programming tasks can be configured with the Devskiller project descriptor file:

1. Create a devskiller.json file.
2. Place it in the root directory of your project.

Here is an example project descriptor:

{
  "verification" : {
    "testNamePatterns" : [".*verify_pack.*"],
    "pathPatterns" : ["**verify_pack**"]
  }
}

You can find more details about the devskiller.json descriptor in our documentation.

Automatic verification with verification tests

The solution submitted by the candidate may be verified using automated tests. You’ll just have to define which tests should be treated as verification tests.

All files classified as verification tests will be removed from the project prior to inviting the candidate.

To define verification tests, you need to set two configuration properties in devskiller.json:

• testNamePatterns - an array of RegEx patterns which should match all the names of the verification tests.
• pathPatterns - an array of GLOB patterns which should match all the files containing verification tests. All the files that match defined patterns will be deleted from candidates' projects and will be added to the projects during the verification phase. These files will not be visible to the candidate during the test.

In our sample project all verification tests are in the VerifyTestClassifier class and the class is located in file which names contains verify_pack. In this case the following patterns will be sufficient:

"testNamePatterns" : [ ".*verify_pack.*" ],
"pathPatterns" : [ "**verify_pack**" ]