Displaying 1 to 7 from 7 results

benchm-ml - A minimal benchmark for scalability, speed and accuracy of commonly used open source implementations (R packages, Python scikit-learn, H2O, xgboost, Spark MLlib etc

  •    R

This project aims at a minimal benchmark for scalability, speed and accuracy of commonly used implementations of a few machine learning algorithms. The target of this study is binary classification with numeric and categorical inputs (of limited cardinality i.e. not very sparse) and no missing data, perhaps the most common problem in business applications (e.g. credit scoring, fraud detection or churn prediction). If the input matrix is of n x p, n is varied as 10K, 100K, 1M, 10M, while p is ~1K (after expanding the categoricals into dummy variables/one-hot encoding). This particular type of data structure/size (the largest) stems from this author's interest in some particular business applications. Note: While a large part of this benchmark was done in Spring 2015 reflecting the state of ML implementations at that time, this repo is being updated if I see significant changes in implementations or new implementations have become widely available (e.g. lightgbm). Also, please find a summary of the progress and learnings from this benchmark at the end of this repo.

useR-machine-learning-tutorial - useR! 2016 Tutorial: Machine Learning Algorithmic Deep Dive http://user2016

  •    Jupyter

Instructions for how to install the necessary software for this tutorial is available here. Data for the tutorial can be downloaded by running ./data/get-data.sh (requires wget). Certain algorithms don't scale well when there are millions of features. For example, decision trees require computing some sort of metric (to determine the splits) on all the feature values (or some fraction of the values as in Random Forest and Stochastic GBM). Therefore, computation time is linear in the number of features. Other algorithms, such as GLM, scale much better to high-dimensional (n << p) and wide data with appropriate regularization (e.g. Lasso, Elastic Net, Ridge).

tgboost - Tiny Gradient Boosting Tree

  •    Java

It is a Tiny implement of Gradient Boosting tree, based on XGBoost's scoring function and SLIQ's efficient tree building algorithm. TGBoost build the tree in a level-wise way as in SLIQ (by constructing Attribute list and Class list). Currently, TGBoost support parallel learning on single machine, the speed and memory consumption are comparable to XGBoost. Handle missing value, XGBoost learn a direction for those with missing value, the direction is left or right. TGBoost take a different approach: it enumerate missing value go to left child, right child and missing value child, then choose the best one. So TGBoost use Ternary Tree.




GBM-perf - Performance of various open source GBM implementations

  •    R

Performance of various open source GBM implementations (h2o, xgboost, lightgbm) on the airline dataset (1M and 10M records). If you don't have a GPU, lightgbm (CPU) trains the fastest.

interpretable_machine_learning_with_python - Practical techniques for interpreting machine learning models

  •    Jupyter

Monotonicity constraints can turn opaque, complex models into transparent, and potentially regulator-approved models, by ensuring predictions only increase or only decrease for any change in a given input variable. In this notebook, I will demonstrate how to use monotonicity constraints in the popular open source gradient boosting package XGBoost to train a simple, accurate, nonlinear classifier on the UCI credit card default data. Once we have trained a monotonic XGBoost model, we will use partial dependence plots and individual conditional expectation (ICE) plots to investigate the internal mechanisms of the model and to verify its monotonic behavior. Partial dependence plots show us the way machine-learned response functions change based on the values of one or two input variables of interest, while averaging out the effects of all other input variables. ICE plots can be used to create more localized descriptions of model predictions, and ICE plots pair nicely with partial dependence plots. An example of generating regulator mandated reason codes from high fidelity Shapley explanations for any model prediction is also presented. The combination of monotonic XGBoost, partial dependence, ICE, and Shapley explanations is likely the most direct way to create an interpretable machine learning model today.

GBM-tune - Tuning GBMs (hyperparameter tuning) and impact on out-of-sample predictions

  •    HTML

The goal of this repo is to study the impact of having one dataset/sample ("the dataset") when training and tuning machine learning models in practice (or in competitions) on the prediction accuracy on new data (that usually comes from a slightly different distribution due to non-stationarity). To keep things simple we focus on binary classification, use only one source dataset with mix of numeric and categorical features and no missing values, we don't perform feature engineering, tune only GBMs with lightgbm and random hyperparameter search (might also ensemble the best models later), and we use only AUC as a measure of accuracy.