Automatically Debugging AutoML Pipelines using Maro: ML Automated Remediation Oracle (Extended Version)

Abstract

Machine learning in practice often involves complex pipelines for data cleansing, feature engineering, preprocessing, and prediction. These pipelines are composed of operators, which have to be correctly connected and whose hyperparameters must be correctly configured. Unfortunately, it is quite common for certain combinations of datasets, operators, or hyperparameters to cause failures. Diagnosing and fixing those failures is tedious and error-prone and can seriously derail a data scientist's workflow. This paper describes an approach for automatically debugging an ML pipeline, explaining the failures, and producing a remediation. We implemented our approach, which builds on a combination of AutoML and SMT, in a tool called Maro. Maro works seamlessly with the familiar data science ecosystem including Python, Jupyter notebooks, scikit-learn, and AutoML tools such as Hyperopt. We empirically evaluate our tool and find that for most cases, a single remediation automatically fixes errors, produces no additional faults, and does not significantly impact optimal accuracy nor time to convergence.

Automatically debugging AutoML pipelines using maro: ML automated remediation oracle

“…The result is a remediated pipeline (8), which the data scientist can inspect directly if they so wish (9). Alternatively, to make the fix easier to understand, the data scientist can send the remediated pipeline and the original pipeline to Maro's explainer component (10). This explains the remediation to the data scientist by rendering it in natural language (11).…”

“…All planned pipelines use common ML operators, mostly from scikitlearn [26], such as LogisticRegression, or OneHotEncoder, but also operators from other scikit-learn compatible libraries, such as a bias mitigator from AIF360 [7] and gradient-boosted trees from LightGBM [21]. The full list of pipelines is available in the extended version of this paper [10].…”

“…There are several categories of error that similar constraints can capture (discussed in more detail in [10]). They are the following (S 2 , S 3 , and S 4 ):…”