litelearn

a python library for building models without fussing over the nitty gritty details for data munging

installation

pip install litelearn

usage

once you have a pandas dataframe you can create a model for your dataset in 3 lines of code:

Regression

# load some dataset
import seaborn as sns

dataset = "penguins"
target = "body_mass_g"
df = sns.load_dataset(dataset).dropna(subset=[target])

# just 3 lines of code to create and evaluate a model
import litelearn as ll

model = ll.regress_df(df, target)
model.display_evaluation() 

Classification

# load some dataset
import seaborn as sns

dataset = "penguins"
target = "species"
df = sns.load_dataset(dataset).dropna(subset=[target])

# just 3 lines of code to create and evaluate a model
import litelearn as ll

model = ll.classify_df(df, target)
model.display_evaluation()

Prediction

prediction is easy too, it will work on any data that resembles the training data. dtypes don't have to match, and you can even have extra columns in your prediction data. missing values and unknown categories will be imputed with the training data's values.

df = ...  # load some dataframe
split = int(len(df) * 0.8)
train_df, val_df = df[:split], df[split:] 
model = ...  # build some model
pred = model.predict(val)  # predict on unseen data

features

• does all the data munging for you, including missing data, categorical data handling
• uses the robust catboost library for gradient boosting, which is known for generating high quality models with little tuning
• supports shap for explainability. call model.display_shap() or model.get_shap() to get the shap values for your model
• supports sklearn's permutation importance call model.display_permutation_importance() or model.get_permutation_importance() to get feature importances that are biased towards the model's performance on test data.
• supports easy pickling: to save your model simply call model.dump("path/to/model.pkl") and to load your model call model.load("path/to/model.pkl")
• for regression models, you can call model.display_residuals() to see the residuals of your model
• it also supports segmeents for your data using the model.set_segments() method. this will create a new column in your dataframe called segment which you can use to group your data. this is useful for seeing how your model performs on different segments of your data.

interactive streamlit demo

To showcase the library, see here for an interactive streamlit web app showing the ability of the library to handle many datasets with ease, including uploading your own dataset.

the code for this demo is on github.com/aviadr1/litelearn-streamlit