def validate_feature_linear(features,
labels,
classes,
n_folds=5,
print_folds=True,
print_absolute=True,
print_logloss=True):
kfold = cv.LabelKFold(labels, n_folds)
model = lda.LDA()
if print_absolute:
score = cross_validation.cross_val_score(model,
features,
classes,
cv=kfold)
if print_absolute: print("absolute scores")
if print_folds: print("\tfolds:", score)
if print_absolute: print("\tmean:", score.mean(), "std:", numpy.std(score))
scores = score_calculation.loglossKFold(features,
classes,
model,
kfold,
given_kfold=True)
if print_logloss: print("logloss scores")
if print_folds: print("\tfolds", scores)
if print_logloss:
print("\tmean:", numpy.mean(scores), "std:", numpy.std(scores))
harald = extractor.calculateDarktoBrightRatio(thumbs[i])
rian = extractor.splitColorFeatures(thumbs[i], splits)
features.append(numpy.append(harald, rian))
#model = grid_search.GridSearchCV(svm.SVC(),{'kernel' : ['poly'], 'C' : [1, 10, 100, 1000], 'degree' : [4,7,10], 'shrinking' : [True, False]})
#model.fit(features, classes)
#print(model.best_estimator_)
#print('\a')
print("Producing KFold indexes")
kfold = cv.KFold(amount, n_folds=5, shuffle=True)
model = lda.LDA()
#model = svm.SVC(kernel = 'linear')
#model = qda.QDA()
score = cross_validation.cross_val_score(model, features, classes, cv=kfold)
print("scores ", score)
print("mean score ", score.mean())
#model = svm.SVC(kernel = 'linear', probability = True)
model = lda.LDA()
#model = neighbors.KNeighborsClassifier(n_neighbors = 1)
scores = score_calculation.loglossKFold(features, classes, model, 5)
print("logloss scores ", scores)
print("logloss score mean ", numpy.mean(scores), " ", numpy.std(scores))
#predictions = cross_validation.cross_val_predict(model, features, classes, cv = kfold)
#wrongIndexes = numpy.nonzero(predictions != classes)
#uniqueWrongs, counts = numpy.unique(numpy.append(predictions[[wrongIndexes]], numpy.array(classes)[[wrongIndexes]]), return_counts = True)
#wrongs = uniqueWrongs[counts > 10]
print('\a')