ariel-data-preprocessing 1.2a1

Signal correction module for Ariel Data Challenge 2025.

Ariel Data Preprocessing

This module contains the FGS1 and AIRS-CH0 signal data preprocessing tools.

Submodules

1. Signal correction (implemented)
2. Signal extraction (partially implemented - AIRS-CH0 data only)

1. Signal correction

Implements the six signal correction steps outline in the Calibrating and Binning Ariel Data notebook shared by the contest organizers.

See the following notebooks for implementation details and plots:

1. Signal correction
2. Signal correction optimization

Example use:

from ariel-data-preprocessing.signal_correction import SignalCorrection

signal_correction = SignalCorrection(
    input_data_path='data/raw',
    output_data_path='data/corrected',
    n_planets=10
)

signal_correction.run()

The signal preprocessing pipeline will write the corrected frames as an HDF5 archive called train.h5 with the following structure:

├── planet_1
|   ├── AIRS-CH0_signal
│   └── FGS1_signal
│
├── planet_1
|   ├── AIRS-CH0_signal
│   └── FGS1_signal
│
.
.
.
└── planet_n
    ├── AIRS-CH0_signal
    └── FGS1_signal

2. Signal extraction

Takes signal corrected data HDF5 output from SignalCorrection().

Selects top n brightest rows of pixels from AIRS-CH0 spectrogram and sums them. Then applies moving average smoothing for each wavelength index across the frames.

See the following notebooks for implementation details and plots:

1. Signal extraction
2. Wavelength smoothing

Example usage:

from ariel-data-preprocessing.signal_correction import SignalExtraction

signal_extraction = SignalExtraction(
    input_data_path='data/corrected',
    output_data_path='data/extracted',
    inclusion_threshold=0.95
)

signal_extraction.run()

Output data will be written to train.h5 in the directory passed to output_data_path. The structure of the HDF5 archive matches the output from SignalCorrection().