lydwhitt-tools 1.1.14

Petrology tools

Lydwhitt_tools

A collection of functions and tools I have developed and find useful during my volcanology PhD.

Whilst I have a lot of tools and functions on my computer that I use regularly, I haven't yet put them all in a place to help others. There are a lot of very simple tasks that waste time, which I've created tools to complete, and I will be adding them to this repository over time.

I started collating this in August 2025 — so bear with me whilst I get it going!

---

Install

pip install lydwhitt-tools

---

Available tools

geoscore_filter(df, phase, smooth_fn=None, smooth_kwargs=None, return_membership=False): End-to-end geochemical filtering pipeline (Liq, Plg, Cpx). Recalculates phase chemistry, fits smooth compositional trends against a progress variable, scores each analysis by how well it follows the main trend (robust local z) and how well it sits within the main high-density KDE modes (HDR membership plus proximity). Returns a continuous geo_score (0 to 1) and a boolean final_pass, plus optional membership diagnostics for debugging and plotting.

filter_fig(df, diagnostics=None, geo_min=None, combine_mode=None, t_col=None, figsize=(18, 15), label_fs=11, tick_fs=9): Diagnostic plot for geoscore_filter results. Builds a multi-panel figure showing the geo_score distribution, the decision space (trend_score vs cluster_score), trend deviation versus the progress variable, retention versus threshold, Harker-style scatter plots with fitted trends, and KDE curves with the kept high-density (HDR) bands.

mahalanobis_filter(df, phase, total_perc=None, percentiles=None) : Two-pass multivariate outlier filter using robust Mahalanobis distance. Filters analyses by oxide totals, computes distances using a Minimum Covariance Determinant estimator, and applies chi-square thresholds to flag outliers in two stages. Returns the original dataframe with pass-1 and pass-2 distances, p-values, and outlier flags merged back for plotting and comparison.

KDE(df, 'column') : This function creates a plottable KDE line for a column of values in a dataframe using the imporved sheather jones method to establish bandwidth. This methodology uses an integrated r script rather than the usual python computing as this is more preferable in geochemical studies. Requires an R installation and the rpy2 Python package.

MD(x, y, z=None) : Returns the x-value at the highest point of a KDE curve (the dominant peak). Optionally set z as a minimum peak height to ignore low-amplitude peaks.

iqr_one_peak(df, 'data', z) : Finds the dominant KDE peak (above a minimum height threshold) and returns that peak location plus a peak-specific IQR (Q1 and Q3) computed only from original data points within the dominant peak window.

recalc(df, phase, anhydrous=True, mol_values=True) : This function calculates the apfu or cation fraction of major elment data for Plg, Cpx, Ol and Liq (WR/Glass/MI) data. anhydrous needs to be specified for Liq data and if you dont want the mol fractions in the final dataframe just add mol_values=true.

recalc_Fe(df) : Ensures a single total-iron column (FeOt) is present by recalculating it from any combination of FeO, Fe2O3, or Fe2O3t (converting Fe2O3 to FeO equivalents). Preserves existing FeOt values and drops raw iron oxide columns.

iron_ratios(df, ratio) : Splits FeOt_Liq into FeO_Liq and Fe2O3_Liq using a supplied Fe3+/FeT ratio. Adds Fe_wt, Fe3_wt, and Fe2_wt intermediate columns and returns the updated dataframe.

---

Worked Examples

Worked examples for each function are provided as Jupyter notebooks on GitHub.
These notebooks show typical inputs, outputs, and common usage patterns.

• geoscore_filter, filter_fig https://shorturl.at/n7TYf

• mahalanobis_filter
  in progress.

• KDE, MD, iqr_one_peak https://shorturl.at/eXg4y

• recalc, recalc_Fe, iron_ratios https://shorturl.at/ZH8vM

---

Features Coming Soon

This repository will grow to include: -data re-oragnisation tools for popular gothermobarometry packages

---

please contact Lydia: [e-mail] with any further questions

License

This project is licensed under the MIT License.