Film Price Guide — Wave Price Predictor

Binary classifier that estimates whether a title’s sold price will rise >10% within 30 days.

Built for Film Price Guide / RetroPrice collector market data.

Model details

Field	Value
Version	`v4`
Algorithm	`RandomForest`
Target	`price_up_10pct_in_30d`
Trained at	`2026-08-15T22:03:38.267530`
Training samples	`24609`
Holdout samples	`6153`
Holdout AUC	`0.6443`
CV AUC (mean)	`0.6618`

Artifacts

wave_prediction_model.pkl — calibrated sklearn classifier
wave_scaler.pkl — StandardScaler fit on training features
model_metrics.json — training metadata + feature list

Features

momentum_7d
momentum_14d
momentum_30d
momentum_60d
momentum_90d
momentum_accel
log_density_7d
log_density_30d
log_listing_count
log_sold_count
ebay_price_velocity
scarcity_score
confidence_score
wave_score
wave_score_delta
listing_growth_rate
log_reddit_mentions
log_reddit_score
graded_count
log_lbxd_fan_count
lbxd_avg_rating
seasonal_multiplier
season_encoded

Top feature importances

momentum_30d: 0.1552
momentum_14d: 0.1405
momentum_60d: 0.1037
momentum_90d: 0.0724
log_lbxd_fan_count: 0.0605
momentum_accel: 0.0599
momentum_7d: 0.0568
lbxd_avg_rating: 0.0553

Intended use

Internal Film Price Guide wave / “heating up” rankings
Research on physical-media sold-price dynamics

Not financial advice. Sold comps are noisy; treat probabilities as directional signals.

How to load

import json, pickle
from huggingface_hub import hf_hub_download

repo = "datamatters24/filmpriceguide-wave-predictor"
model = pickle.load(open(hf_hub_download(repo, "wave_prediction_model.pkl"), "rb"))
scaler = pickle.load(open(hf_hub_download(repo, "wave_scaler.pkl"), "rb"))
metrics = json.load(open(hf_hub_download(repo, "model_metrics.json")))

Sync

Uploaded by scripts/sync_model_to_hf.py from the RetroPrice training pipeline. Last card generated: 2026-08-21 01:00 UTC

Downloads last month: -