abcard

abcard is a Python package primarily used for developing binary classification models and generating reports for production work.
It supports Logit and LGBM models, can generate PDF model reports and production-ready deployment code.
It emphasizes concise and efficient API calls, rich visualizations, and retention of the development process.

Installation

abcard requires Python 3.9 or later, install using pip with:

pip install abcard

Additional dependencies: numpy, pandas, tqdm, statsmodels, scikit-learn, matplotlib, PyMuPDF, lightgbm

Main Features

from abcard import Frame, LogitFrame, LGBMCFrame, Report, ModReport, LogitReport, LGBMCReport

train: 'pandas.DataFrame datasets'
test: 'pandas.DataFrame datasets'
oot: 'pandas.DataFrame datasets'
flag: 'target label (y)'
time: 'name of the time column, optional'
exclude: 'column names to be excluded, optional'
Mod: 'Logit, LGBM models'

df = LogitFrame(flag = flag, time = time, exclude = exclude) # Initial sample set field configuration.
df = LGBMCFrame(flag = flag, time = time, exclude = exclude) # Initial sample set field configuration.
df.set_samp(train, 'train') # Set the sample dataset.
df.get_samp(test, 'test') # Get the sample dataset.
df.del_samp(oot, 'oot') # Delete the sample dataset.

df.describe_sample() # Descriptive analysis of samples.
df.describe_feature() # Descriptive analysis of fearures.
df.chi2_split() # Perform chi-square binning on all features.

df.mergebins() # Merge bins manually.

df.drop_nan(nan = 0.9) # Various feature filtering methods starting with 'drop'.

df.transform() # Convert the sample set into bins or WOE.

train_set = df.get_xy(label = 'train') # Get the X and y for model training or prediction.
df.get_metric() # Retrieve the model's evaluation metrics on all sample sets.
df.scorecard() # Calculate Logistic Regression Scorecard for Selected Features.

df._mod = Mod # A fully trained model.

rep = LGBMCReport(df) # Initialize a PyMuPDF Document object for a LGBMClassifier model Frame object.
rep.design()
rep.describe_feature()
rep.bins() # Generate a chapter for the feature binning results.
rep.filter() # Generate a chapter for the feature filtering results.
rep.modres() # Generate a chapter containing the model results.
rep.analysis('train') # Generate a chapter containing the model analysis and evaluation for the specified samples.
rep.plotcuts(cores = 4) # Generate a chapter containing the model binning plots.
rep.code() # Generate a chapter containing the deployment code.
rep.log()
rep.save("model_name_report.pdf")
rep.close()

License and Copyright

abcard is available under open-source AGPL and commercial license agreements.