Deschide articolul original într-un tab nou Schimbarea acestei opțiuni se aplică imediat și reîncarcă pagina.
LECTURA
← Înapoi la Urmăritori

Thierry Moudiki’s webpage

thierrymoudiki.github.io/feed.xml
Vizitează site-ul

Cel mai recent articol

Ultima verificare

Urmăritori 0 urmăritori
Articole: 17

  1. Choosing the size of the calibration set in conformal prediction with ahead::conformalize()

    (Thierry Moudiki’s webpage)

    In version 0.39.0 of the ahead package, the conformalize() function has been updated to include a new parameter called split_prob.

  2. Machine-learning loss reserving vs Mack Chain-Ladder under settlement speed-up and calendar inflation (Python version)

    (Thierry Moudiki’s webpage)

    Claude was the copilot for this post about my Techtonique python package mlreserving, and yes, we’re cooked (no, actually, we’ll have to be very smart, competent, and creative).

  3. mlreserve: machine-learning loss reserving on ‘real-world’ triangles (based on a ‘ChainLadder’ fork)

    (Thierry Moudiki’s webpage)

    This post presents a function mlReserve() for machine-learning loss reserving on ‘real-world’ triangles, based on a fork of R’s ‘ChainLadder’ package (https://github.com/thierrymoudiki/ChainLadder/tree/master).

  4. Analytics Pipeline for Dashboards, with Python, R and Javascript

    (Thierry Moudiki’s webpage)

    I had a lot of fun this morning, brainstorming and assembling this Analytics Pipeline for Dashboards with Claude (and yes it takes much, much more than only 10 prompts).

  5. Python version of Semi-parametric option pricing based on underlying’s historical data (accepted at the osQF 2026 (ex R/Finance) conference)

    (Thierry Moudiki’s webpage)

    Python implementation of “Semi-parametric option pricing based on underlying’s historical data” (accepted as osQF (ex R/Finance) 2026, https://www.researchgate.net/publication/414513095_Semi-parametric_option_pricing_based_on_underlying’s_historical_data).

  6. Semi-parametric option pricing based on underlying’s historical data (accepted at the osQF 2026 (ex R/Finance) conference)

    (Thierry Moudiki’s webpage)

    This post is a follow-up to my previous posts on semi-parametric option pricing.

  7. Model-agnostic prediction intervals in Python and R: does nnetsauce’s `QuantileRegressor` hold up?

    (Thierry Moudiki’s webpage)

    Model-agnostic prediction intervals in Python and R: does nnetsauce’s QuantileRegressor hold up?

  8. ahead (Time Series Forecasting with uncertainty quantification) gets a lot faster to install: most dependencies are now optional

    (Thierry Moudiki’s webpage)

    If you’ve used ahead before, you may remember install.packages("ahead") or pip install ahead taking a while, since it dragged in two dozen modeling packages up front regardless of which one you actually planned to use.

  9. Skip the R/Python runtime: fast tabular dashboards with Observable Framework

    (Thierry Moudiki’s webpage)

    If you’ve built dashboards with Shiny or Streamlit, you know the pattern: a Python or R process runs on a server, holds your data in memory, and re-executes app logic on every interaction.

  10. PCARVFL vs CTGAN for synthetic tabular data generation on an insurance pricing dataset

    (Thierry Moudiki’s webpage)

    This notebook applies PCARVFLSimulator (from the Python package synthe) — a GAN-like tabular data synthesizer built from PCA scores, a Random Vector Functional-Link (RVFL) network, and residual bootstrapping — to the French Motor Third-Party Liability (fr…

  11. ‘Zero-Shot Probabilistic Stock Returns Forecasting with Pretrained RVFL Networks’ accepted at COPA 2026 (and to appear in the Proceedings of Machine Learning Research)

    (Thierry Moudiki’s webpage)

    As for last year, I’ve got a poster accepted at COPA conference.

  12. ‘PCARVFLSimulator’: a GAN-like tabular data synthesizer built from PCA scores, a Random Vector Functional-Link network, and residuals bootstrapping

    (Thierry Moudiki’s webpage)

    Generative Adversarial Networks are the default reach for synthetic tabular data, but they are not the only route to the same destination.

  13. ‘garchf’: GARCH probabilistic forecasting with package ‘forecast’-style interface (and ‘rugarch’ under the hood)

    (Thierry Moudiki’s webpage)

    The package garchf wraps the extensive GARCH modeling machinery in rugarch behind a forecast-style interface, so that GARCH-based forecasts can be produced, plotted, and evaluated with the same conventions used throughout the forecast ecosystem (in the sp…

  14. GPopt for R: Bayesian and conformal optimization of black-box functions and hyperparameter tuning

    (Thierry Moudiki’s webpage)

    This post describes the R version of Python’s GPopt (https://docs.techtonique.net/GPopt/GPopt.html), a package for ‘Bayesian’ optimization of black-box functions and machine learning hyperparameter tuning, using Gaussian Process Regression and other confo…

  15. My last R posts: How conformalization helps weak models, fast conformal prediction with jackknife+ (and no refitting), and sklearn in R

    (Thierry Moudiki’s webpage)

    This post is mainly (but not only) a test, because I had a broken xml feed for my R posts, and I wanted to see if it was fixed.

  16. Natively Interpretable Boosting

    (Thierry Moudiki’s webpage)

    Natively Interpretable Boosting Gradient boosted decision trees are accurate but usually opaque: to explain a prediction you rely on a post hoc method (SHAP, LIME, permutation importance) that approximates the model after the fact.

  17. Fast conformal prediction (no refitting) for some Machine Learning models via closed-form jackknife plus

    (Thierry Moudiki’s webpage)

    It’s surprisingly fast to obtain conformal jackknife+ prediction intervals for Machine Learning models of the form \(\hat{y} = Sy\) (including Ordinary Least Squares, Ridge Regression, Random Vector Functional Link Networks, Kernel Ridge Regression, smoot…