Forecasting¶

Performance snapshots for this task, comparing deep-river models against strong baselines on standard river datasets.

AirlinePassengers (target only)¶

Summary¶

Model	MAE	RMSE	SMAPE	Memory in Mb	Time in s
Deep River GRU	294.115	315.123	197.554	0.0329781	18.9884
Deep River LSTM	295.086	316.16	198.842	0.03298	13.5827
Deep River Linear	57.66	72.2396	33.7468	0.0158186	3.74214
Deep River Liquid	293.711	314.679	197.037	0.0328302	32.3297
Deep River MLP	67.8867	82.534	40.9045	0.0372353	6.61261
Deep River N-BEATS	48.5183	60.8551	21.0501	0.0746565	11.7495
Deep River RNN	294.655	315.604	198.381	0.0330992	11.1271
Holt-Winters	39.4407	50.5302	13.2046	0.0044384	0.566215
SNARIMAX	32.7727	41.7392	11.378	0.00846672	1.02697

Charts¶

TrumpApproval (target only)¶

Summary¶

Model	MAE	RMSE	SMAPE	Memory in Mb	Time in s
Deep River GRU	21.7723	23.6523	81.9672	0.0334435	106.264
Deep River LSTM	22.4099	24.1704	84.935	0.0334454	71.471
Deep River Linear	1.33074	4.49723	4.06335	0.0162611	18.1434
Deep River Liquid	21.9987	23.7984	82.7777	0.0332499	201.501
Deep River MLP	1.82508	5.54169	5.73855	0.0376472	33.3228
Deep River N-BEATS	1.06645	2.67359	2.80861	0.0751066	64.0395
Deep River RNN	22.1253	24.0073	83.9181	0.0335493	59.1101
Holt-Winters	0.393099	0.517923	0.970281	0.00443077	0.721287
SNARIMAX	0.310209	0.414372	0.765805	0.00873375	2.70793

Charts¶

Datasets¶

AirlinePassengers (target only)

Monthly number of international airline passengers.

The stream contains 144 items and only one single feature, which is the month. The goal is to predict the number of passengers each month by capturing the trend and the seasonality of the data.

Name  AirlinePassengers                                                                                                          
Task  Regression

Samples 144
Features 1
Sparse False
Path /Users/cedrickulbach/Documents/Projects/deep-river/.venv/lib/python3.12/site-packages/river/datasets/airline-passengers.csv

Forecasting benchmark variant: only the historical target values are used.

TrumpApproval (target only)

Donald Trump approval ratings.

This dataset was obtained by reshaping the data used by FiveThirtyEight for analyzing Donald Trump's approval ratings. It contains 5 features, which are approval ratings collected by 5 polling agencies. The target is the approval rating from FiveThirtyEight's model. The goal of this task is to see if we can reproduce FiveThirtyEight's model.

Name  TrumpApproval                                                                                                             
Task  Regression

Samples 1,001
Features 6
Sparse False
Path /Users/cedrickulbach/Documents/Projects/deep-river/.venv/lib/python3.12/site-packages/river/datasets/trump_approval.csv.gz

Forecasting benchmark variant: only the historical target values are used.

Models¶

Holt-Winters

HoltWinters (
  alpha=0.3
  beta=0.1
  gamma=0.1
  seasonality=12
  multiplicative=False
)

SNARIMAX

SNARIMAX (
  p=1
  d=1
  q=0
  m=12
  sp=1
  sd=0
  sq=0
  regressor=Pipeline (
    steps=OrderedDict({'StandardScaler': StandardScaler (
  with_std=True
), 'LinearRegression': LinearRegression (
  optimizer=SGD (
    lr=Constant (
      learning_rate=0.01
    )
  )
  loss=Squared ()
  l2=0.
  l1=0.
  intercept_init=0.
  intercept_lr=Constant (
    learning_rate=0.01
  )
  clip_gradient=1e+12
  initializer=Zeros ()
)})
  )
)

Deep River Linear

LinearForecaster (
  n_features=0
  window_size=12
  loss_fn="mse"
  optimizer_fn="sgd"
  lr=0.005
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River MLP

MLPForecaster (
  n_features=0
  window_size=12
  n_width=16
  n_layers=2
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River RNN

RNNForecaster (
  n_features=0
  window_size=12
  hidden_size=32
  num_layers=1
  nonlinearity="tanh"
  dropout=0.
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River GRU

GRUForecaster (
  n_features=0
  window_size=12
  hidden_size=32
  num_layers=1
  dropout=0.
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River LSTM

LSTMForecaster (
  n_features=0
  window_size=12
  hidden_size=32
  num_layers=1
  dropout=0.
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River Liquid

LiquidForecaster (
  n_features=0
  window_size=12
  hidden_size=32
  num_layers=1
  dropout=0.
  time_delta=1.
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Deep River N-BEATS

NBEATSForecaster (
  n_features=0
  window_size=12
  n_width=32
  n_layers=2
  n_blocks=2
  loss_fn="mse"
  optimizer_fn="adam"
  lr=0.001
  is_feature_incremental=False
  device="cpu"
  seed=42
  gradient_clip_value=1.
)

Environment¶

Python implementation: CPython
Python version       : 3.12.13
IPython version      : 9.6.0

river       : 0.25.0
numpy       : 2.5.0
scikit-learn: 1.5.2
pandas      : 2.2.3
scipy       : 1.16.2

Compiler    : Clang 22.1.3 
OS          : Linux
Release     : 6.17.0-1018-azure
Machine     : x86_64
Processor   : x86_64
CPU cores   : 4
Architecture: 64bit