A first lab, start to finish

This tutorial runs one lab with a class, from choosing a lesson to reading the grades. It uses Lesson 1, Two moons, because it needs no download and trains in seconds. Everything in it is free, in every edition, except grading the class with spl grade at the end, which needs an Education license.

Before the class

Try the lesson yourself. Choose Help › Lessons…, then Lesson 1: Two moons. The lesson’s starter becomes a project of your own and opens on the Lab page: the lab’s title, its checks and their points, the blocks it locks, the worksheet, and Check my work.
Read the worksheet on the Lab page. It says the goal, the steps, the questions and what to hand in. Students see the same worksheet beside their model.
Do the lab as a student would. Train the starter, press Check my work, add the hidden layer the worksheet asks for, train again and check again. It takes about 30 minutes the first time and 10 the second.
Check the lab’s computers. On one of them, run spl doctor. Each line it prints is a check, with what to do when one fails. Running a lab covers the computers in more detail.

In the class

Students open the lesson the same way: Help › Lessons…, then the lesson. Each gets a project of their own, named after the lesson. Starting the lesson again makes a new project, so a student who wants to start over loses nothing.
Point them to the Lab page. The worksheet is there, beside the checks. Most questions in a first lab are answered by its steps.
Check my work marks the model as it is on the screen, saved or not, with the rules the grader uses. Each check says Passed or Not passed, its points and, when it fails, the lab’s feedback. Encourage students to press it often: it trains nothing and answers at once.
Locked blocks (here, the input, the output and the training blocks) show a lock. A student can move them, but not change them, so every hand-in has the same inputs and outputs.
Training is the student’s evidence, not the grade. The checks look at the model and its settings, never at a training run, so a class gets the same marks on any computer.

Handing in

A student hands in with File › Export…, then Project file. That makes one .speliora file with the model, its training settings, the lab and the student’s session summary. It holds no datasets and no runs, so it is small enough for any learning management system.

Collect the files the way your school collects any assignment, such as an upload to the LMS, then download them all into one folder.

Grading

An Education license on the computer that grades lets you mark the whole folder at once:

run spl grade submissions/ --assignment assignment.json -o grades/, with the lesson’s own assignment.json from the repository (content/lessons/L01-two-moons/);
grades/grades.csv has one row per file: its points, each check passed or failed, the errors its model still had, the feedback, and the session’s times;
grades/summary.txt says what the class got right and wrong, such as how many hit each error;
grades/feedback/ has one text file per student, which you can hand back as it is.

Without a license, spl grade reads nothing and says so. You can still open each hand-in in the app and press Check my work, which gives the same marks.

After the class

Start the debrief from the summary. The most common error code is usually the best first question for the class.
Read the session times. The summary’s median time to a first model without errors and to full points tells you whether the lab fitted its slot.
Next: Running a lab covers the computers and the network; Writing a lab covers labs of your own.