4. Non Parameters Scenarios#

When dealing with new reasearch problems, it is common to have new scenarios that are not covered by the current quantifiers. In this case, there are problems when:

  • Learner (e.g. classifier) does not have fit, predict and predict_proba methods, or it is an external learner that does not have a sklearn interface

  • You already have the predictions of the mid task (e.g. classifier) and you don’t want to remake the predictions.

The solution for this is to use the set_arguments method to set the arguments of the quantifier.

Note

Recommended Usage

When using the set_parameters, you don’t need to pass any learner when instantiating the quantifier, but you still need to use the fit and predict methods of the quantifier.

from mlquantify import set_arguments
from mlquantify.methods import EMQ
from sklearn.ensemble import RandomForestClassifier
import numpy as np

X_train = np.random.rand(100, 10)
y_train = np.random.randint(0, 2, 100)
X_test = np.random.rand(10, 10)
y_test = np.random.randint(0, 2, 10)

model = RandomForestClassifier()
model.fit(X_train, y_train)

# parameters
y_pred = model.predict(X_test) # predictions of the test set
posteriors_train = model.predict_proba(X_train) # predictions of the training set generated via cross validation
posteriors_test = model.predict_proba(X_test) # predictions of the test set
y_labels = y_train # Generated via cross validation
y_pred_train = model.predict(X_train) # predictions of the training set generated via cross validation

set_arguments(y_pred=y_pred,
            posteriors_train=posteriors_train,
            posteriors_test=posteriors_test,
            y_labels=y_labels,
            y_pred_train=y_pred_train)

quantifier = EMQ()
quantifier.fit(X_train, y_train)
pred = quantifier.predict(X_test)
print(pred)

Each argument is optinal depending on the quantifier you are using. The arguments are:

  • y_pred: Array-like

    • Predictions for the test set,

    • must be the same length as the test set.

    • Must be python list, numpy array or pandas series.

  • posteriors_train: Array-like

    • Class probabilities for the training set computed via cross-validation.

    • Must be of shape (n_samples, n_classes), with each row summing to 1.0.

    • Must be python list, numpy array or pandas Dataframe.

  • posteriors_test: Array-like

    • Class probabilities for the test set.

    • Must be of shape (n_samples_test, n_classes), with each row summing to 1.0.

    • Must be python list, numpy array or pandas Dataframe.

  • y_labels: Array-like

    • Ground truth labels for the training set used during cross-validation.

    • Must be the same length as the posteriors_train parameter.

    • Must be python list, numpy array or pandas series.

  • y_pred_train: Array-like

    • Predictions for the training set generated through cross-validation.

    • Must be the same length as the posteriors_train parameter.

    • Must be python list, numpy array or pandas series.

This Page