6

I am doing a shap tutorial, and attempting to get the shap values for each person in a dataset

from sklearn.model_selection import train_test_split
import xgboost
import shap
import numpy as np
import pandas as pd
import matplotlib.pylab as pl

X,y = shap.datasets.adult()
X_display,y_display = shap.datasets.adult(display=True)

# create a train/test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=7)
d_train = xgboost.DMatrix(X_train, label=y_train)
d_test = xgboost.DMatrix(X_test, label=y_test)
params = {
    "eta": 0.01,
    "objective": "binary:logistic",
    "subsample": 0.5,
    "base_score": np.mean(y_train),
    "eval_metric": "logloss"
}
#model = xgboost.train(params, d_train, 5000, evals = [(d_test, "test")], verbose_eval=100, early_stopping_rounds=20)

xg_clf = xgboost.XGBClassifier()
xg_clf.fit(X_train, y_train)
explainer = shap.TreeExplainer(xg_clf, X_train)
#shap_values = explainer(X)
shap_values = explainer.shap_values(X)

going through the Python3 interpreter, shap_values is a massive array of 32,561 persons, each with a shap value for 12 features.

For example, the first individual has the following SHAP values:

>>> shap_values[0]
array([ 0.76437867, -0.11881508,  0.57451954, -0.41974955, -0.20982443,
       -0.38079952, -0.00986504,  0.32272505, -3.04392116,  0.00411322,
       -0.26587735,  0.02700199])

However, which value applies to which feature is a complete mystery to me.

the documentation says:

For models with a single output this returns a matrix of SHAP values
        (# samples x # features). Each row sums to the difference between the model output for that
        sample and the expected value of the model output (which is stored in the expected_value
        attribute of the explainer when it is constant). For models with vector outputs this returns
        a list of such matrices, one for each output

When I go to explainer which produced shap_values I see that I can get feature names:

explainer.data_feature_names
['Age', 'Workclass', 'Education-Num', 'Marital Status', 'Occupation', 'Relationship', 'Race', 'Sex', 'Capital Gain', 'Capital Loss', 'Hours per week', 'Country']

but I cannot see how to get feature names within shap_values at the Python interpreter, if they're even there:

>>> shap_values.
shap_values.all(           shap_values.compress(      shap_values.dump(          shap_values.max(           shap_values.ravel(         shap_values.sort(          shap_values.tostring(
shap_values.any(           shap_values.conj(          shap_values.dumps(         shap_values.mean(          shap_values.real           shap_values.squeeze(       shap_values.trace(
shap_values.argmax(        shap_values.conjugate(     shap_values.fill(          shap_values.min(           shap_values.repeat(        shap_values.std(           shap_values.transpose(
shap_values.argmin(        shap_values.copy(          shap_values.flags          shap_values.nbytes         shap_values.reshape(       shap_values.strides        shap_values.var(
shap_values.argpartition(  shap_values.ctypes         shap_values.flat           shap_values.ndim           shap_values.resize(        shap_values.sum(           shap_values.view(
shap_values.argsort(       shap_values.cumprod(       shap_values.flatten(       shap_values.newbyteorder(  shap_values.round(         shap_values.swapaxes(      
shap_values.astype(        shap_values.cumsum(        shap_values.getfield(      shap_values.nonzero(       shap_values.searchsorted(  shap_values.T              
shap_values.base           shap_values.data           shap_values.imag           shap_values.partition(     shap_values.setfield(      shap_values.take(          
shap_values.byteswap(      shap_values.diagonal(      shap_values.item(          shap_values.prod(          shap_values.setflags(      shap_values.tobytes(       
shap_values.choose(        shap_values.dot(           shap_values.itemset(       shap_values.ptp(           shap_values.shape          shap_values.tofile(        
shap_values.clip(          shap_values.dtype          shap_values.itemsize       shap_values.put(           shap_values.size           shap_values.tolist(

My primary question: How can I figure out which feature in

['Age', 'Workclass', 'Education-Num', 'Marital Status', 'Occupation', 'Relationship', 'Race', 'Sex', 'Capital Gain', 'Capital Loss', 'Hours per week', 'Country']

applies to which number in each row of shap_values?

>>> shap_values[0]
array([ 0.76437867, -0.11881508,  0.57451954, -0.41974955, -0.20982443,
       -0.38079952, -0.00986504,  0.32272505, -3.04392116,  0.00411322,
       -0.26587735,  0.02700199])

I would assume that the features are in the same order, but I have no evidence for that.

My secondary question: how can I find the feature names in shap_values?

Lucas
  • 6,869
  • 5
  • 29
  • 44
con
  • 5,767
  • 8
  • 33
  • 62

2 Answers2

4

The features are indeed in the same order, as you assume; see how to extract the most important feature names? and how to get feature names from explainer issues in Github.

To find the feature name, you simply need to access the element with the same index of the array with the names

For example:

shap_values = np.array([
    0.76437867, -0.11881508,  0.57451954, -0.41974955, -0.20982443,
   -0.38079952, -0.00986504,  0.32272505, -3.04392116,  0.00411322,
   -0.26587735,  0.02700199])
features_names = ['Age', 'Workclass', 'Education-Num', 'Marital Status', 'Occupation',
                  'Relationship', 'Race', 'Sex', 'Capital Gain', 'Capital Loss',
                  'Hours per week', 'Country']

features_names[shap_values.argmin()]  # the index 8 -> Capital Gain
features_names[shap_values.argmax()]  # the index 0 -> Age
oustella
  • 63
  • 5
Lucas
  • 6,869
  • 5
  • 29
  • 44
  • [This answer](https://github.com/slundberg/shap/issues/632#issuecomment-823923453) in the first link you posted works great. – Jakob Dec 01 '22 at 18:08
1

If you find this answer useful, upvote @lucas's answer and GitHub user ba1mn's post. I'm just adding it here in case the link breaks.

The following function will return the feature names along with their corresponding importance in a DataFrame.

def global_shap_importance(model, X):
    """ Return a dataframe containing the features sorted by Shap importance
    Parameters
    ----------
    model : The tree-based model 
    X : pd.Dataframe
         training set/test set/the whole dataset ... (without the label)
    Returns
    -------
    pd.Dataframe
        A dataframe containing the features sorted by Shap importance
    """
    explainer = shap.Explainer(model)
    shap_values = explainer(X)
    cohorts = {"": shap_values}
    cohort_labels = list(cohorts.keys())
    cohort_exps = list(cohorts.values())
    for i in range(len(cohort_exps)):
        if len(cohort_exps[i].shape) == 2:
            cohort_exps[i] = cohort_exps[i].abs.mean(0)
    features = cohort_exps[0].data
    feature_names = cohort_exps[0].feature_names
    values = np.array([cohort_exps[i].values for i in range(len(cohort_exps))])
    feature_importance = pd.DataFrame(
        list(zip(feature_names, sum(values))), columns=['features', 'importance'])
    feature_importance.sort_values(
        by=['importance'], ascending=False, inplace=True)
    return feature_importance
Jakob
  • 663
  • 7
  • 25