How oiibar/animescoreforecast works
This repository is a research project and local application for predicting anime scores from pre-release metadata and synopsis text. The user-facing workflow collects title attributes and synopsis in Streamlit, builds model features, applies saved prediction models, and displays a bounded score. Separate research stages evaluate feature engineering, target-tail strategies, supervised models and ensembles, and unsupervised analyses. Data collection and shared preprocessing support those workflows. Some stage implementation is only partially sampled; edges are limited to imports, calls, and README-documented relationships visible in the evidence.
Main entry points: exp_common.py Shared experiment data · features.py Structured features · app.py Streamlit predictor · e3_models.py Model comparisons
Components 16
Data collection
- Score-band collector
scripts/collect_data.py - Raw snapshot
data/raw/anime_raw.json
Shared data and features
- Clean dataset
data/processed/anime_clean.csv - Shared experiment data
src/exp_common.py - Structured features
src/features.py - Synopsis cleaning and embeddings
src/text_features.py
Model research
- Feature experiments
experiments/e1_features.py - Target-tail experiments
experiments/e2_target.py - Model comparisons
experiments/e3_models.py - Unsupervised analysis
experiments/e4_unsupervised.py - Final model checks
experiments/v1_check.py
Prediction application
- Saved prediction models
models/final_model.joblib - Streamlit predictor
app/app.py - Predicted score
app/app.py
Other components
- User
- AniList API
Connections 20
- Score-band collector → AniList API: queries
scripts/collect_data.py - Score-band collector → Raw snapshot: writes snapshot
scripts/collect_data.py - Streamlit predictor → Shared experiment data: loads data
app/app.py - Streamlit predictor → Structured features: builds input
app/app.py - Structured features → Shared experiment data: uses preprocessing
src/features.py - Structured features → Synopsis cleaning and embeddings: cleans synopsis
src/features.py - Shared experiment data → Synopsis cleaning and embeddings: imports cleaner
src/exp_common.py - Feature experiments → Shared experiment data: uses shared data
experiments/e1_features.py - Feature experiments → Structured features: tests features
experiments/e1_features.py - Target-tail experiments → Shared experiment data: runs validation
experiments/e2_target.py - Model comparisons → Shared experiment data: runs comparisons
experiments/e3_models.py - Model comparisons → Structured features: uses feature prep
experiments/e3_models.py - Unsupervised analysis → Shared experiment data: runs analysis
experiments/e4_unsupervised.py - Final model checks → Model comparisons: imports models
experiments/v1_check.py - Final model checks → Structured features: checks features
experiments/v1_check.py - User → Streamlit predictor: enters inputs
README.md - Streamlit predictor → Saved prediction models: predicts with
- Streamlit predictor → Predicted score: displays score
app/app.py - Raw snapshot → Clean dataset: prepared dataset
- Clean dataset → Shared experiment data: loads dataset