Introduction

TableMage is a Python package for low-code/conversational clinical data science.

Domain expertise is extremely valuable in data science and quantitative research. However, experts in fields such as medicine and healthcare often lack the necessary statistical and computational experience to analyze data on their own.

TableMage simplifies data science, making it accessible to researchers with little to no quantitative background. TableMage empowers domain experts to perform their own data analysis and machine learning modeling without needing to learn the intricacies of data science (Python, pandas, scikit-learn, ML theory). TableMage’s Python API is designed to be intuitive and easy to use, with low-code functions that abstract away the complexity of data science.

Below, we list the steps of a typical data science workflow and describe how TableMage’s low-code API can simplify each step.

Data Loading and Cleaning

Data must be extracted and cleaned prior to analysis. TableMage does not handle data loading and cleaning. TableMage works with wide format tabular data in the form of a pandas DataFrame. See tablemage.Analyzer for details.

Data Exploration

New datasets need to be explored prior to further processing and modeling. Data scientists working with Python typically use pandas and matplotlib/seaborn to explore data.

TableMage simplifies the data exploration process by providing low-code tools for the following tasks:

1. Summary statistics

2. Statistical testing

3. Data visualization

See tablemage.Analyzer.eda() for details.

Data Preprocessing

Before modeling, data must be preprocessed (e.g., missing values imputed, categorical variables encoded, etc.).

TableMage streamlines the data preprocessing process, reducing several minutes of documentation reading and 10+ lines of code down to zero minutes of documentation reading and one line of code.

See tablemage.Analyzer.impute(), tablemage.Analyzer.scale(), tablemage.Analyzer.drop_highly_missing_vars(), tablemage.Analyzer.dropna(), and tablemage.Analyzer.onehot() for details.

Note

Categorical variables are automatically one-hot encoded in the modeling process if you do not one-hot encode them beforehand. Also, observations with missing values are automatically dropped in the modeling process.

Statistical Modeling

Statistical modeling is sometimes preferred over machine learning modeling because it provides interpretable results.

TableMage improves the statistical modeling process by providing low-code functions for linear regression analysis.

See tablemage.Analyzer.ols() and tablemage.Analyzer.logit() for details.

Machine Learning Modeling

Machine learning modeling is often preferred over statistical modeling because it can handle more complex relationships in the data.

TableMage simplifies the machine learning modeling process by providing low-code functions for classification and regression models.

See tablemage.Analyzer.classify() and tablemage.Analyzer.regress() for details. The different models available are listed in Machine Learning Models (tm.ml).