List of classification methods

[Boblancvich]

Hi,
Where would I be able to find a full list of Classification methods that River covers?

Thank you!

[MaxHalford]

Hey @Boblancvich!

Alas we don't have a nice list in our docs. But you can do this:

import importlib
import inspect
from river import base

for submodule in importlib.import_module("river.api").__all__:

    def is_classifier(obj):
        return inspect.isclass(obj) and issubclass(obj, base.Classifier)

    for name, obj in inspect.getmembers(
        importlib.import_module(f"river.{submodule}"), is_classifier
    ):
        print(f"{submodule}.{name}")

base.Classifier
base.MiniBatchClassifier
compat.SKL2RiverClassifier
dummy.NoChangeClassifier
dummy.PriorClassifier
ensemble.ADWINBaggingClassifier
ensemble.AdaBoostClassifier
ensemble.AdaptiveRandomForestClassifier
ensemble.BaggingClassifier
ensemble.LeveragingBaggingClassifier
ensemble.SRPClassifier
ensemble.StackingClassifier
ensemble.VotingClassifier
facto.FFMClassifier
facto.FMClassifier
facto.FwFMClassifier
facto.HOFMClassifier
imblearn.HardSamplingClassifier
imblearn.RandomOverSampler
imblearn.RandomSampler
imblearn.RandomUnderSampler
linear_model.ALMAClassifier
linear_model.LogisticRegression
linear_model.PAClassifier
linear_model.Perceptron
linear_model.SoftmaxRegression
model_selection.SuccessiveHalvingClassifier
multiclass.OneVsOneClassifier
multiclass.OneVsRestClassifier
multiclass.OutputCodeClassifier
multioutput.ClassifierChain
multioutput.MonteCarloClassifierChain
multioutput.ProbabilisticClassifierChain
naive_bayes.BernoulliNB
naive_bayes.ComplementNB
naive_bayes.GaussianNB
naive_bayes.MultinomialNB
neighbors.KNNClassifier
tree.ExtremelyFastDecisionTreeClassifier
tree.HoeffdingAdaptiveTreeClassifier
tree.HoeffdingTreeClassifier
tree.LabelCombinationHoeffdingTreeClassifier
tree.SGTClassifier

I hope that helps!

[Dennis1989]

@smastelini In general, yes, it is a good idea! SBERT especially performs super well, at least for traditional clustering. For stream clustering, I never considered them as inference is slow. Additionally, language models are not really updated (also word embeddings), which is quite problematic in our domain because of concept drift. I played around with incremental embeddings, which are promising but still WIP...

[Boblancvich]

Great, thanks for the advice guys :)

[Boblancvich]

Hi again,

I was also curious as to why there is no Online version of SVM?

Thanks!

[MaxHalford]

Good question. Actually, there is. If you use the logistic regression with a Hinge loss, that results in a linear SVM. Then, if you need to handle non-linearity, then it's recommended to use the RBFSampler class from the feature_extraction module :)

[Boblancvich]

Hi @MaxHalford ,

Thanks for this idea! I'll give it a go!