Explore anime data with statistically robust rankings.
AnimeAnalyst pulls titles from MyAnimeList (Jikan v4), can cross-check AniList, lets you filter by time/type/genre, and plots top-K charts using a Bayesian score with a tunable prior (α).
It’s a lightweight, reproducible tool for data-minded anime fans, students, and creators.

Why this project?
Public scores are noisy—small shows can look “too good,” big shows are penalized by variance.
AnimeAnalyst applies Bayesian shrinkage to stabilize rankings, exposes the α (prior strength), and keeps a CSV cache so you can iterate without hammering APIs.

How it works (30s):

1. Fetch from Jikan (/anime) with server-side filters.
2. Flatten results → CSV cache.
3. (Optional) Pull AniList and map MAL↔AniList.
4. Apply local filters (type/status/year/votes/genres).
5. Compute Bayesian score: (\frac{n}{n+m}\bar{s} + \frac{m}{n+m}C).
6. Plot clean horizontal bar charts.

• 🔎 Jikan v4 fetch with filters: q, type, status, min_score, date range, sfw, pagination.
• 🧹 Local filters: type, status, year range, minimum voters, genres (any/all) by name.
• 🎭 Genre helper: genre_all to list all MAL genres; genre_any <names|ids> accepts names or numeric IDs.
• 🧮 Bayesian score using MAL score + scored_by with a robust prior (default max(1000, median_votes)).
• 📊 Matplotlib horizontal bar charts for the top-K titles.
• 💾 CSV cache (default data/anime_cache.csv) to avoid refetching.
• 🔌 Hooks to merge AniList data and compute a fused score.

# Python 3.10+ recommended
pip install -r requirements.txt

## Run

From the project root:

```bash
python -m anime_analyst.cli

This starts an interactive shell. Type parameters line by line, then start to run.

General:

• help – show all parameters and examples
• show – print current parameter values
• reset – reset to defaults
• quit / exit – leave the program

Genre helpers:

• genre_all – list all MAL anime genres as id : name

• genre_any <tokens> – set “match any” genres. tokens can be names or IDs, separated by space or comma

  • genre_any Action, Fantasy
  • genre_any 1 10

Booleans toggle if value omitted:

• sfw → toggles True/False
• no_fetch → toggles True/False

Finally:

• start – fetch (unless no_fetch), filter, score, and plot

Any param can be given as key value or key=value. Repeating a key overwrites the previous value.

Top TV airing now, at least 50k votes, any of Action/Adventure:

genre_all                     # see all available genres (id : name)
genre_any Action, Adventure   # or: genre_any 1 2 (ids also OK)
type tv
status airing
min_scored_by 50000
max_pages 3
start

Best movies 2010–2020 with SFW only:

type movie
year_from 2010
year_to 2020
min_scored_by 10000
sfw
start

Reuse cached CSV only (no network):

no_fetch
start

Bayesian score:

$$ \text{score} = \frac{n}{n+m},\bar{s} + \frac{m}{n+m},C $$

• $\bar{s}$: MAL mean score

• $n$: MAL scored_by

• $C$: global prior (vote-weighted average across filtered rows)

• $m$: prior weight

  • default: m = max(1000, median(scored_by))
  • override with prior_m

Shrinkage intuition: the effective “alpha” is $\alpha = \frac{m}{n+m}$. Larger $m$ ⇒ stronger pull toward $C$.

• MAL via Jikan v4

  • Endpoint: https://api.jikan.moe/v4/anime
  • Server-side params we use: q, type, status, start_date, end_date, min_score, sfw, page, limit
  • We respect HTTP 429 and Retry-After.

• AniList (optional)

  • anime_analyst/clients/anilist.py includes a GraphQL client (ID/title search).
  • anime_analyst/core/merge.py shows how to map MAL ↔ AniList and (optionally) compute a fused score.

Genres are resolved via https://api.jikan.moe/v4/genres/anime. The CLI command genre_all lists them; genre_any accepts names or IDs (we convert IDs → names for local filtering).

The following sample figures live in the repo under assets/ as p1.png … p5.png. p1 and p2 use the same time window but different Bayesian priors (hence different effective α / shrinkage).

src/
  anime_analyst/
    cli.py                   # interactive CLI (entry point)
    clients/
      jikan.py               # Jikan fetch + flatten
      anilist.py             # AniList GraphQL client (optional enrichment)
    core/
      filter.py              # local filters (type/status/year/votes/genres)
      scoring.py             # bayesian_score + compute_bayesian_scores
      plotting.py            # plot_hbar_top
      merge.py               # MAL↔AniList merge + fusion logic (optional)
    data/
      genre.py               # GenreResolver (name/id mapping, genre_all/any)
      io.py                  # CSV save/load
assets/
  p1.png p2.png p3.png p4.png p5.png

Run as a module:

python -m anime_analyst.cli

requirements.txt:

requests>=2.31,<3
matplotlib>=3.8,<3.10
numpy>=1.24,<3
# rich>=13.7   # (optional) pretty CLI tables

Install:

pip install -r requirements.txt

• By default, fetched rows are flattened and saved to data/anime_cache.csv.
• Set no_fetch to reuse the CSV without network calls.

• No module named 'anime_analyst' Run from the project root with module mode:

  python -m anime_analyst.cli

  Or mark src/ as Sources Root / set PYTHONPATH to the project root.

• HTTP 429 We back off using Retry-After and a small delay. Lower max_pages / limit_per_page if needed.

• IDE files keep conflicting Ensure .gitignore contains:

  .idea/
  __pycache__/
  *.pyc
  .venv/
  data/*.csv
  

  And remove tracked IDE files:

  git rm -r --cached .idea

MIT

::contentReference[oaicite:0]{index=0}