Publications by T. Moudiki
A deeper learning architecture in nnetsauce
As you may know already (or not), nnetsauce contains CustomClassifier and CustomRegressor models, which allow to derive a new statistical/Machine Learning (ML) model from another one.Creating this new ML model is achieved by doing some feature engineering. That is, by augmenting the original set of explanatory variables, using random or quasirand...
2216 sym Python (1659 sym/7 pcs) 10 img
Explaining xgboost predictions with the teller
Nowadays, explaining the decisions of Statistical/Machine learning (ML) algorithms is becoming a must, and also, mainstream. In healthcare for example, ML explainers could help in understanding how black-box – but accurate – ML prognosis about patients are formed.One way to obtain these explanations (here is another way that I introduced in a...
3043 sym Python (1187 sym/5 pcs) 2 img
Compatibility of nnetsauce and mlsauce with scikit-learn
Disclaimer: I have no affiliation with the scikit-learn team.Thanks to inheritance, nnetsauce and mlsauce models share a lot of properties with scikit-learn’s Statistical/Machine learning (ML) models. That’s to say: if you’re already familiar with scikit-learn, you won’t have to spend a lot of time figuring out how do nnetsauce and mlsauc...
2234 sym Python (5320 sym/16 pcs) 2 img
Bayesian Optimization with GPopt
Due to the way it mixes several – relatively – simple concepts, Bayesian optimization (BO) is one of the most elegant mathematical tool I’ve encountered so far. In this post, I introduce GPopt, a tool for BO that I implemented in Python (no technical docs yet, but coming soon). The examples of GPopt’s use showcased here are based on Gaus...
2687 sym Python (3234 sym/26 pcs) 8 img
Bayesian Optimization with GPopt Part 2 (save and resume)
Two weeks ago, I presented GPopt: a Python package for Bayesian optimization. In particular, I’ve presented a way to stop the optimizer and resume it later by adding more iterations.This week, I present a way to save and resume, that makes the optimizer’s data persistent. Behind this saving feature, are hidden Python shelves which are – so...
996 sym Python (1674 sym/7 pcs) 4 img
Hyperparameters tuning with GPopt
Statistical/Machine learning models can have multiple hyperparameters that control their performance (out-of-sample accuracy, area under the curve, Root Mean Squared Error, etc.). In this post, in order to determine these hyperparameters for mlsauce’s LSBoostClassifier (on the wine dataset), cross-validation is used along with a Bayesian optimi...
1072 sym Python (2913 sym/9 pcs)
Classification using linear regression
In this post, I illustrate classification using linear regression, as implemented in Python/R package nnetsauce, and more precisely, in nnetsauce’s MultitaskClassifier. If you’re not interested in reading about the model description, you can jump directly to the 2nd section, “Two examples in Python”. In addition, the source code is relati...
3487 sym Python (2771 sym/6 pcs)
Tuning and interpreting LSBoost
There is a plethora of Automated Machine Learning tools in the wild, implementing Machine Learning (ML) pipelines from data cleaning to model validation. In this post, the input data set is already cleaned and pre-processed (diabetes dataset); the ML model is already chosen too, mlsauce’s LSBoost. We are going to focus on two important steps of...
2061 sym Python (4572 sym/15 pcs) 4 img
Forecasting with `ahead` (Python version)
A few weeks ago, I introduced the R version of ahead, a package for univariate and multivariate time series forecasting. A Python version, built on top of the R version, is now available on PyPI and GitHub. Here is how to install it:1st method: from PyPI (stable version) pip install ahead ...
273 sym Python (20 sym/1 pcs)
Explaining a Keras _neural_ network predictions with the-teller
Last year, in a previous post, I’ve used Python package the-teller to explain an xgboost model’s predictions. After reading today’s post, you’ll be able to use that same package, the-teller, to explain predictions of a Keras neural network trained on tabular data.We start by installing the following tools:An AutoML system based on Keras: ...
635 sym Python (51 sym/2 pcs)