Data

…

🔒 Coming Soon — Public Release with Paper Publication
…

Dataset Overview

…

Time Span
2010 – 2025
Samples
2.1M
Features
35 / 145
Cycles
24 + 25
…

Specifications

Source NASA SDO / HMI  ·  SHARP CEA data product
Time span 2010-05 — 2025
Cadence 12 minutes
Base parameters 25 raw SHARP parameters (see BASE_FEATS_25.txt)
Engineered features …
Label …
Split protocol Stratified Cycle Splitting (chronological, AR-disjoint)
File format .parquet (Apache Parquet, snappy compression)
Version v1.0-preview

Split Protocol Schematic

…

Files

…

File Size
df_train.parquet
Training set — used for model fitting (Cycle 24/25 chronological, AR-disjoint)
~ 1.1 GB Request
df_val.parquet
Validation set — used for early stopping and threshold tuning
~ 160 MB Request
df_test.parquet
Test set — held out, used only for final reporting (no leakage)
~ 180 MB Request
BASE_FEATS_25.txt
25 raw SHARP parameter names (before feature engineering)
< 1 KB Request
selected_features_35.json
35 final features used by the model (Random-Forest selected)
< 5 KB Request
split_info.json
AR-level split metadata (which AR went to train / val / test)
< 50 KB Request
README.md
Documentation, usage instructions, and citation guidelines
< 10 KB Request
📦 Get the entire dataset in one click
…
Request Access

…

Quick Start

Once you have the files, load and explore the dataset in a few lines:

import pandas as pd
import json
# 1. Load splits
df_train = pd.read_parquet('df_train.parquet')
df_val   = pd.read_parquet('df_val.parquet')
df_test  = pd.read_parquet('df_test.parquet')
# 2. Load feature list
with open('selected_features_35.json') as f:
    feats = json.load(f)
X_train, y_train = df_train[feats], df_train['label']
print(f'Train: {len(X_train):,} samples, positive rate: {y_train.mean():.3f}')

Reproducing the published benchmark

…

License & Citation

License

…

…

Cite this dataset

…

…

Contact & Support

…