causalicp.Result

The result of running causalicp.fit() is returned in a causalicp.Result object, which contains the estimate, accepted sets, p-values, etc.

class causalicp.Result(target, data, estimate, accepted, rejected, conf_intervals, set_pvalues, set_coefs)

The result of running Invariant Causal Prediction, produced as output of causalicp.fit().

p

The total number of variables in the data (including the response/target).

Type

int

target

The index of the response/target.

Type

int

estimate

The estimated parental set returned by ICP, or None if all sets of predictors were rejected.

Type

set or None

accepted_sets

A list containing the accepted sets of predictors.

Type

list of set

rejected_sets

A list containing the rejected sets of predictors.

Type

list of set

pvalues

A dictionary containing the p-value for the causal effect of each individual predictor. The target/response is included in the dictionary and has value nan.

Type

dict of (int, float)

conf_intervals

A 2 x p array of floats representing the confidence interval for the causal effect of each variable. Each column corresponds to a variable, and the first and second row correspond to the lower and upper limit of the interval, respectively. The column corresponding to the target/response is set to nan.

Type

numpy.ndarray or None

Example

>>> import causalicp as icp
>>> result = icp.fit(data, 3)

>>> result.p
4

>>> result.target
3

>>> result.estimate
set()

>>> result.accepted_sets
[{1}, {2}, {0, 1}, {1, 2}, {0, 1, 2}]

>>> result.rejected_sets
[set(), {0}, {0, 2}]

>>> result.pvalues
{0: 1, 1: 0.187430598304751, 2: 1, 3: nan}

>>> result.conf_intervals
array([[0.        , 0.        , 0.        ,        nan],
       [2.37257655, 1.95012059, 5.88760917,        nan]])

When all sets are rejected (e.g. there is a model violation), the estimate and confidence intervals are set to None:

>>> result = icp.fit(data_bad_model, 3)
>>> result.estimate
>>> result.conf_intervals

And the individual p-value for the causal effect of each variable is set to 1:

>>> result.pvalues
{0: 1, 1: 1, 2: 1, 3: nan}