Multiclass classification

Hyperplane (limited 5000)

Summary

Model	Accuracy	MicroF1	MacroF1	Memory in Mb	Time in s
Deep River LSTM	0.8882	0.8882	0.888163	0.048192	1192.18
Deep River Logistic	0.9084	0.9084	0.908396	0.0271883	157.744
Deep River MLP	0.5004	0.5004	0.498917	0.049367	291.074
Logistic regression	0.9108	0.9108	0.9108	0.00967312	36.7864
[baseline] Last Class	0.503301	0.503301	0.503278	0.000510216	14.4975
[baseline] Prior Class	0.494699	0.494699	0.49095	0.000611305	14.912

Charts

LED (limited 5000)

Summary

Model	Accuracy	MicroF1	MacroF1	Memory in Mb	Time in s
Deep River LSTM	0.338	0.338	0.328551	0.03547	1147.47
Deep River Logistic	0.3606	0.3606	0.31801	0.0215616	148.098
Deep River MLP	0.1048	0.1048	0.0525165	0.0437403	279.831
Deep River RNN	0.3524	0.3524	0.313258	0.0357409	433.301
Logistic regression	0.0928	0.0928	0.016987	0.00505733	18.9079
[baseline] Last Class	0.0980196	0.0980196	0.0975498	0.00121212	7.60275
[baseline] Prior Class	0.105421	0.105421	0.0468459	0.00116062	8.76123

Charts

RandomRBF (limited 5000)

Summary

Model	Accuracy	MicroF1	MacroF1	Memory in Mb	Time in s
Deep River LSTM	0.5368	0.5368	0.489181	0.0331888	1075.22
Deep River Logistic	0.5066	0.5066	0.314396	0.0202723	146.062
Deep River MLP	0.3482	0.3482	0.112211	0.0424776	282.092
Deep River RNN	0.5136	0.5136	0.353501	0.0334597	423.671
Logistic regression	0.3628	0.3628	0.157195	0.00439358	17.8173
[baseline] Last Class	0.276855	0.276855	0.243844	0.000563622	7.77983
[baseline] Prior Class	0.34907	0.34907	0.13395	0.000718117	8.54999

Charts

Datasets

Hyperplane (limited 5000)

Hyperplane(limited n=5000)

LED (limited 5000)

LED(limited n=5000)

RandomRBF (limited 5000)

RandomRBF(limited n=5000)

Models

Logistic regression

Pipeline (
  StandardScaler (
    with_std=True
  ),
  LogisticRegression (
    optimizer=SGD (
      lr=Constant (
        learning_rate=0.005
      )
    )
    loss=Log (
      weight_pos=1.
      weight_neg=1.
    )
    l2=0.
    l1=0.
    intercept_init=0.
    intercept_lr=Constant (
      learning_rate=0.01
    )
    clip_gradient=1e+12
    initializer=Zeros ()
  )
)

Deep River Logistic

Pipeline (
  StandardScaler (
    with_std=True
  ),
  LogisticRegressionInitialized (
    n_features=10
    n_init_classes=2
    loss_fn="cross_entropy"
    optimizer_fn="sgd"
    lr=0.005
    output_is_logit=True
    is_feature_incremental=True
    is_class_incremental=True
    device="cpu"
    seed=42
    gradient_clip_value=None
  )
)

Deep River MLP

Pipeline (
  StandardScaler (
    with_std=True
  ),
  MultiLayerPerceptronInitialized (
    n_features=10
    n_width=5
    n_layers=5
    n_init_classes=2
    loss_fn="cross_entropy"
    optimizer_fn="sgd"
    lr=0.005
    output_is_logit=True
    is_feature_incremental=True
    is_class_incremental=True
    device="cpu"
    seed=42
    gradient_clip_value=None
  )
)

Deep River LSTM

Pipeline (
  StandardScaler (
    with_std=True
  ),
  LSTMClassifier (
    n_features=10
    hidden_size=32
    n_init_classes=2
    loss_fn="cross_entropy"
    optimizer_fn="adam"
    lr=0.001
    output_is_logit=True
    is_feature_incremental=True
    is_class_incremental=True
    device="cpu"
    seed=42
    gradient_clip_value=None
  )
)

Deep River RNN

Pipeline (
  StandardScaler (
    with_std=True
  ),
  RNNClassifier (
    n_features=10
    hidden_size=32
    num_layers=1
    nonlinearity="tanh"
    n_init_classes=2
    loss_fn="cross_entropy"
    optimizer_fn="adam"
    lr=0.001
    output_is_logit=True
    is_feature_incremental=True
    is_class_incremental=True
    device="cpu"
    seed=42
    gradient_clip_value=None
  )
)

[baseline] Last Class

NoChangeClassifier ()

[baseline] Prior Class

PriorClassifier ()

Environment

Python implementation: CPython
Python version       : 3.12.12
IPython version      : 9.6.0

river       : 0.22.0
numpy       : 1.26.4
scikit-learn: 1.5.2
pandas      : 2.2.3
scipy       : 1.16.2

Compiler    : Clang 21.1.4 
OS          : Linux
Release     : 6.11.0-1018-azure
Machine     : x86_64
Processor   : x86_64
CPU cores   : 4
Architecture: 64bit