Usage: EDA

Once a TransformedOutcome() object has been instantiated, pylift offers a couple methods for simple feature EDA: up.NIV() (Net Information Value) and up.NWOE() (Net Weight of Evidence).

See Data Exploration with Weight of Evidence and Information Value in R for more details.

Net Weight of Evidence

Weight of Evidence comes from a simple Bayesian decomposition of relative lift as a function of features:

\[\log \frac{P(Y = 1|X_j)}{P(Y = 0|X_j)} = \log \frac{P(Y=1)}{P(Y=0)} + \underbrace{\log \frac{P(X_j | Y=1)}{P(X_j | Y =0)}}_{\text{Weight of Evidence}}\]

Net Weight of Evidence (NWOE) is the difference in Weight of Evidence (WOE) between the treatment and control groups.

\[\text{NWOE} = \text{WOE}_t - \text{WOE}_c\]

Where Weight of Evidence is defined as:

\[\text{WOE}_{ij} = \log \frac{P(X_j \elem B_i | Y = 1)}{P(X_j \elem B_i | Y = 0)}\]

where \(B_i\) indicates a bin \(i\), and the subscript \(j\) indicates a particular feature.

This can be accessed in pylift as follows:

up = TransformedOutcome(df)
up.NWOE()

up.NWOE() can take two arguments: n_bins (the number of bins) and feats_to_use (a subset of features over which to calculate NWOE).

The base routine can be accessed from the pylift.explore module (from pylift.explore.base import _NWOE).

Net Information Value

Information Value is the sum of all WOE values, weighted by the absolute difference in the numerator and denominator.

\[\text{IV}_j = \int \log \frac{P(X_j | Y = 1)}{P(X_j | Y=0)} (P(X_j | Y = 1) - P(X_j | Y = 0)) \ dx\]

For the net weight of evidence, the numerator and denominator can be rewritten as:

\[\text{NWOE_j} = \log \frac{P(x_j | Y = 1, W = 1) P(x_j | Y = 0, W= 0)}{P(x_j | Y =1, W=0) P(x_j | Y = 0, W = 1)}\]

Net information value, then, sums the NWOE values, weighted by the difference between the above numerator and denominator.

This can be accessed in pylift as follows:

up = TransformedOutcome(df)
up.NIV()

up.NIV() accepts feats_to_use and n_bins, as it requires NWOE to be calculated as a pre-requisite. But up.NIV() also accepts n_iter – up.NIV() will bootstrap the training set to obtain error bars, and n_iter specifies the number of iterations to use.

The base routine can be accessed from the pylift.explore module (from pylift.explore.base import _NIV).