Skip to content

Create a study (no code)

This walks through the Administration UI to create and run a study without writing any code, using the built-in fastcompare template.

Prerequisites

  • EasyStudy running (see Quickstart) and datasets fetched.
  • An admin account (/signup → /login).

Step 1 — start a new fastcompare study

From Administration, click Create under fastcompare. You get a parameter form; every field has a ? help tooltip.

Step 2 — choose data & elicitation

  • Data loader — the domain/dataset participants will see (e.g. MovieLens Latest Small (demo) for a quick first try, or Goodbooks-10k, Filtered ML-25M).
  • Preference elicitation — how you gather initial signal to personalize. Popularity Sampling works in the lightweight core and needs no extras.

Step 3 — choose algorithms to compare

  • Pick 2–3 algorithms (e.g. EASE + a baseline). Their parameters appear as editable fields automatically (this is the annotation-driven UI — e.g. EASE exposes l2 and positive_threshold).
  • Only algorithms whose dependencies are installed appear. Install the recbole / tensorflow / lenskit extra (uv sync --extra recbole, etc. — see Concepts) to unlock more.

Step 4 — configure the flow

  • Recommendation size (K) — items shown per algorithm per round.
  • Number of iterations (N) — how many comparison rounds.
  • Layout — rows vs columns (affects presentation/position bias).
  • Prompts / texts — the wording participants see.
  • Prolific code (optional) — if recruiting via Prolific.

Step 5 — create & wait for initialization

Submit. The study initializes in the background (training the algorithms on the dataset). When it finishes it flips to active and shows a join URL. If something fails, the error surfaces in the admin UI (initialization_error).

Step 6 — distribute the join URL

Share the join URL with participants (or paste it into Prolific/e-mail). Each participant goes through consent → elicitation → N rounds → final page/questionnaire.

Step 7 — monitor & collect

Watch progress and open Results for built-in metrics, or query the database directly (Data model).

Creating variants

Create multiple instances from the same template with different parameters (e.g. rows vs columns, different algorithms) — each gets its own URL and its own data. This is the standard way to run between-subject conditions today.

Coming in v2

A drag-and-drop no-code study designer (interleave questionnaires, attention checks, branching) and a template gallery are the headline v2 features — this page will grow with them.