hazykiller/trappist1-dip-atlas

---
license: cc-by-4.0
---
# TRAPPIST-1 Dip Atlas

This dataset contains an automatically extracted and cleaned set of **brightness dips** from the TESS light curve of **TRAPPIST-1**, an ultracool dwarf star with 7 Earth-sized planets.

The goal is to provide a **machine-learning-ready dataset** of dip segments for tasks like:
- transit vs noise classification
- anomaly detection
- clustering of transit morphologies
- testing generative / denoising models

## Files

- `trappist1_dip_catalog.csv`
A table where each row corresponds to one dip segment.

Columns (may vary slightly depending on version):

- `dip_index` – index of the dip segment
- `orig_time_index` – index in the original light curve where the dip is centered
- `hdbscan_cluster` – unsupervised cluster label (HDBSCAN on PCA/UMAP space)
- `kmeans_cluster` – KMeans cluster label in PCA space
- `min_flux` – minimum normalized flux in the segment
- `median_flux` – mean or median flux level of the segment
- `depth` – `median_flux - min_flux` (approximate depth of the dip)
- `duration_points` – number of time samples significantly below the median

- `trappist1_dip_flux_cleaned.npy`
A NumPy array of shape `(n_dips, n_samples)` containing the cleaned and normalized dip segments.

- Each row = one dip
- Each column = one time step within the fixed-size window
- Values are normalized such that typical out-of-transit flux is ~1.0

- `trappist1_dip_umap_2d.npy`
2D UMAP embedding of each dip segment, shape `(n_dips, 2)`.

This is useful for:
- visualization of clusters in 2D
- downstream clustering / anomaly detection in a compact space.

## Usage

```python
import numpy as np
import pandas as pd

catalog = pd.read_csv("trappist1_dip_catalog.csv")
dip_flux = np.load("trappist1_dip_flux_cleaned.npy")
dip_umap = np.load("trappist1_dip_umap_2d.npy")

print(catalog.head())
print(dip_flux.shape, dip_umap.shape)