def fit_predict(self,
Gtr,
Ytr,
Gt,
Yt,
grid_search,
acc_param="F1Mean",
RS=0,
VERBOSE=False):
"""Learn the model on training dataset and predict on the testing dataset.
Input:
- Gtr (array): training subset
- Ytr (array): true labels of training subset
- Gt (array): testing subset
- Yt (array): true labels of testing subset
- grid_search (dict): grid search for the CV
- acc (str): accuracy parameter for the cross-validation (default "F1Mean", otherwise it will be the OA)
- RS (int) : random seed used for the stratified k-fold CV
- VERBOSE (bool): verbose (default: False)
Return:
- confMatrix (object ConfusionMatrix): confusion matrix issued from the classification
- yp (array): vector of predicted labels of the testing subset
- grid.best_params_ (dict): combination of parameters that gave the best results during the CV
TODO: implement own scoring parameter
"""
if acc_param == "F1Mean":
score = 'f1_macro'
else:
score = 'accuracy'
## Initialization of the stratifield K-CV
cv = StratifiedKFold(Ytr, n_folds=self.n_folds, random_state=RS)
#Implementation of a fit and a predict methods with parameters from grid search
grid = GridSearchCV(self.pipe,
param_grid=grid_search,
scoring=score,
verbose=VERBOSE,
cv=cv,
n_jobs=3)
grid.fit(Gtr, Ytr)
model = grid.best_estimator_
if VERBOSE:
print grid.best_score_
#Learn model
model.fit(Gtr, Ytr) #could use refit in version 0.19 of sklearn
#Predict
yp = model.predict(Gt)
#Compute confusion matrix
confMatrix = ConfusionMatrix()
confMatrix.compute_confusion_matrix(yp, Yt)
return confMatrix, yp, grid.best_params_
