pca.py

import h5py
import numpy as np
import polars as pl
from scipy.signal import find_peaks
from sklearn.decomposition import PCA

with h5py.File("data/periodograms.h5", "r") as f:
    tic = f["meta/tic_id"][:]
    sec = f["meta/sector"][:]
    freq = f["frequency"][:]
    power = f["power"][:]

ids = list(zip(tic, sec))

results = []
n = 10
for p in power:
    peaks, _ = find_peaks(p, distance=500)
    top_peaks = peaks[np.argsort(p[peaks])[-n:]]
    
    peak_powers = (p[top_peaks] - np.mean(p))/np.std(p)
    peak_freqs = (freq[top_peaks]) / np.max(freq)
    results.append(np.column_stack((peak_freqs, peak_powers)).ravel())

X = np.array(results)
pca = PCA(n_components=2)
X_pca = pca.fit_transform(X)

alfven_df = pl.read_csv("data/targets_unc.csv")
df = pl.DataFrame({"tic_id": tic, "sector": sec, "PC1": X_pca[:, 0], "PC2": X_pca[:, 1]})
df = df.join(alfven_df, on=["tic_id", "sector"])
df.write_csv("data/results.csv")