scorecardpy

The woebin function is used to create bins based on Weight of Evidence. You can adjust these bins either interactively or by providing a manual dictionary of breaks.

To apply manual breaks, pass a dictionary to the breaks_list parameter in sc.woebin. The keys should be the column names, and the values should be lists of break points.

Once bins are defined, use sc.woebin_ply to transform your training and testing DataFrames into their corresponding WOE values.

# Generate initial bins
bins = sc.woebin(dt_s, y="creditability")

# Define manual breaks for specific columns
breaks_adj = {
    'age.in.years': [26, 35, 40],
    'other.debtors.or.guarantors': ["none", "co-applicant%,%guarantor"]
}

# Re-run binning with manual breaks
bins_adj = sc.woebin(dt_s, y="creditability", breaks_list=breaks_adj)

# Transform data into WOE values
train_woe = sc.woebin_ply(train, bins_adj)
test_woe = sc.woebin_ply(test, bins_adj)

A typical end-to-end workflow using scorecardpy follows these steps:

1. Prepare Data: Load data and filter variables using sc.var_filter.
2. Split Data: Use sc.split_df to create training and testing sets.
3. Binning: Perform WOE binning with sc.woebin and transform data with sc.woebin_ply.
4. Modeling: Train a logistic regression model (e.g., using sklearn.linear_model.LogisticRegression) on the WOE-transformed training data.
5. Evaluation: Evaluate performance using sc.perf_eva and check stability with sc.perf_psi.
6. Scaling: Generate a scorecard using sc.scorecard and apply it to raw data using sc.scorecard_ply to get final credit scores.

import scorecardpy as sc
from sklearn.linear_model import LogisticRegression

# 1. Data Preparation
dat = sc.germancredit()
dt_s = sc.var_filter(dat, y="creditability")
train, test = sc.split_df(dt_s, 'creditability').values()

# 2. WOE Binning
bins_adj = sc.woebin(dt_s, y="creditability")
train_woe = sc.woebin_ply(train, bins_adj)
test_woe = sc.woebin_ply(test, bins_adj)

y_train = train_woe.loc[:,'creditability']
X_train = train_woe.loc[:,train_woe.columns != 'creditability']
y_test = test_woe.loc[:,'creditability']
X_test = test_woe.loc[:,train_woe.columns != 'creditability']

# 3. Logistic Regression
lr = LogisticRegression(penalty='l1', C=0.9, solver='saga', n_jobs=-1)
lr.fit(X_train, y_train)

# 4. Performance Evaluation
train_pred = lr.predict_proba(X_train)[:,1]
test_pred = lr.predict_proba(X_test)[:,1]
sc.perf_eva(y_train, train_pred, title = "train")
sc.perf_eva(y_test, test_pred, title = "test")

# 5. Scorecard Scaling
card = sc.scorecard(bins_adj, lr, X_train.columns)
train_score = sc.scorecard_ply(train, card, print_step=0)
test_score = sc.scorecard_ply(test, card, print_step=0)

# 6. PSI
sc.perf_psi(
  score = {'train':train_score, 'test':test_score},
  label = {'train':y_train, 'test':y_test}
)