LSTM Regression Model¶

In [1]:

from deep_river.regression import RollingRegressor
from river import (
    metrics,
    compose,
    preprocessing,
    datasets,
    stats,
    feature_extraction,
)
from torch import nn
from tqdm import tqdm

from deep_river.regression import RollingRegressor from river import ( metrics, compose, preprocessing, datasets, stats, feature_extraction, ) from torch import nn from tqdm import tqdm

In [2]:

def get_hour(x):
    x["hour"] = x["moment"].hour
    return x

def get_hour(x): x["hour"] = x["moment"].hour return x

Simple RNN Regression Model¶

In [3]:

class RnnModule(nn.Module):

    def __init__(self, n_features, hidden_size):
        super().__init__()
        self.n_features = n_features
        self.rnn = nn.RNN(
            input_size=n_features, hidden_size=hidden_size, num_layers=1
        )
        self.fc = nn.Linear(in_features=hidden_size, out_features=1)

    def forward(self, X, **kwargs):
        output, hn = self.rnn(X)  # lstm with input, hidden, and internal state
        return self.fc(output[-1, :])

class RnnModule(nn.Module): def __init__(self, n_features, hidden_size): super().__init__() self.n_features = n_features self.rnn = nn.RNN( input_size=n_features, hidden_size=hidden_size, num_layers=1 ) self.fc = nn.Linear(in_features=hidden_size, out_features=1) def forward(self, X, **kwargs): output, hn = self.rnn(X) # lstm with input, hidden, and internal state return self.fc(output[-1, :])

In [4]:

dataset = datasets.Bikes()
metric = metrics.MAE()

model_pipeline = compose.Select(
    "clouds", "humidity", "pressure", "temperature", "wind"
)
model_pipeline += get_hour | feature_extraction.TargetAgg(
    by=["station", "hour"], how=stats.Mean()
)
model_pipeline |= preprocessing.StandardScaler()
model_pipeline |= RollingRegressor(
    module=RnnModule,
    loss_fn="mse",
    optimizer_fn="sgd",
    window_size=20,
    lr=1e-2,
    hidden_size=32,  # parameters of MyModule can be overwritten
    append_predict=True,
)
model_pipeline

dataset = datasets.Bikes() metric = metrics.MAE() model_pipeline = compose.Select( "clouds", "humidity", "pressure", "temperature", "wind" ) model_pipeline += get_hour | feature_extraction.TargetAgg( by=["station", "hour"], how=stats.Mean() ) model_pipeline |= preprocessing.StandardScaler() model_pipeline |= RollingRegressor( module=RnnModule, loss_fn="mse", optimizer_fn="sgd", window_size=20, lr=1e-2, hidden_size=32, # parameters of MyModule can be overwritten append_predict=True, ) model_pipeline

Out[4]:

['clouds', [...]

Select ( clouds humidity pressure temperature wind )

get_hour

def get_hour(x): x["hour"] = x["moment"].hour return x

y_mean_by_station_and_hour

TargetAgg ( by=['station', 'hour'] how=Mean () target_name="y" )

StandardScaler

StandardScaler ( with_std=True )

RollingRegressor

RollingRegressor ( module=None loss_fn="mse" optimizer_fn="sgd" lr=0.01 is_feature_incremental=False window_size=20 append_predict=True device="cpu" seed=42 )

In [5]:

for x, y in tqdm(dataset.take(5000)):
    y_pred = model_pipeline.predict_one(x)
    metric.update(y_true=y, y_pred=y_pred)
    model_pipeline.learn_one(x=x, y=y)
print(f"MAE: {metric.get():.2f}")

for x, y in tqdm(dataset.take(5000)): y_pred = model_pipeline.predict_one(x) metric.update(y_true=y, y_pred=y_pred) model_pipeline.learn_one(x=x, y=y) print(f"MAE: {metric.get():.2f}")

0it [00:00, ?it/s]

1it [00:00,  1.32it/s]

43it [00:00, 67.61it/s]

95it [00:00, 150.91it/s]

145it [00:01, 222.99it/s]

191it [00:01, 276.40it/s]

240it [00:01, 328.17it/s]

290it [00:01, 372.09it/s]

341it [00:01, 407.35it/s]

392it [00:01, 434.36it/s]

441it [00:01, 444.48it/s]

489it [00:01, 441.17it/s]

536it [00:01, 435.48it/s]

585it [00:01, 448.57it/s]

636it [00:02, 465.84it/s]

685it [00:02, 471.82it/s]

733it [00:02, 471.79it/s]

783it [00:02, 478.74it/s]

832it [00:02, 475.08it/s]

881it [00:02, 478.90it/s]

930it [00:02, 474.18it/s]

978it [00:02, 473.61it/s]

1026it [00:02, 472.70it/s]

1076it [00:03, 478.32it/s]

1126it [00:03, 482.34it/s]

1175it [00:03, 482.14it/s]

1224it [00:03, 482.02it/s]

1273it [00:03, 480.28it/s]

1323it [00:03, 485.48it/s]

1374it [00:03, 491.39it/s]

1424it [00:03, 485.53it/s]

1473it [00:03, 469.39it/s]

1523it [00:03, 476.88it/s]

1572it [00:04, 479.50it/s]

1622it [00:04, 484.55it/s]

1671it [00:04, 485.63it/s]

1720it [00:04, 483.89it/s]

1770it [00:04, 487.62it/s]

1821it [00:04, 492.89it/s]

1872it [00:04, 496.10it/s]

1922it [00:04, 493.71it/s]

1972it [00:04, 485.80it/s]

2022it [00:04, 488.00it/s]

2072it [00:05, 490.82it/s]

2122it [00:05, 493.50it/s]

2172it [00:05, 489.03it/s]

2221it [00:05, 477.90it/s]

2270it [00:05, 480.45it/s]

2320it [00:05, 483.41it/s]

2369it [00:05, 479.67it/s]

2417it [00:05, 478.69it/s]

2465it [00:05, 478.17it/s]

2515it [00:05, 484.47it/s]

2564it [00:06, 482.40it/s]

2614it [00:06, 485.44it/s]

2664it [00:06, 486.28it/s]

2713it [00:06, 477.68it/s]

2761it [00:06, 476.44it/s]

2813it [00:06, 486.90it/s]

2862it [00:06, 471.59it/s]

2910it [00:06, 470.15it/s]

2958it [00:06, 462.33it/s]

3008it [00:07, 472.57it/s]

3058it [00:07, 478.48it/s]

3106it [00:07, 473.88it/s]

3157it [00:07, 482.38it/s]

3206it [00:07, 477.33it/s]

3255it [00:07, 480.77it/s]

3307it [00:07, 489.95it/s]

3357it [00:07, 491.79it/s]

3410it [00:07, 500.45it/s]

3461it [00:07, 491.46it/s]

3512it [00:08, 496.57it/s]

3562it [00:08, 497.41it/s]

3613it [00:08, 498.96it/s]

3663it [00:08, 495.46it/s]

3713it [00:08, 485.39it/s]

3763it [00:08, 488.13it/s]

3813it [00:08, 491.18it/s]

3865it [00:08, 497.26it/s]

3915it [00:08, 492.44it/s]

3965it [00:08, 479.16it/s]

4014it [00:09, 477.70it/s]

4064it [00:09, 484.02it/s]

4115it [00:09, 489.03it/s]

4166it [00:09, 494.26it/s]

4216it [00:09, 486.31it/s]

4268it [00:09, 494.77it/s]

4319it [00:09, 497.11it/s]

4369it [00:09, 497.06it/s]

4419it [00:09, 485.99it/s]

4469it [00:10, 487.42it/s]

4520it [00:10, 492.12it/s]

4572it [00:10, 497.62it/s]

4622it [00:10, 495.40it/s]

4672it [00:10, 487.50it/s]

4722it [00:10, 489.70it/s]

4774it [00:10, 497.22it/s]

4826it [00:10, 502.29it/s]

4877it [00:10, 500.93it/s]

4928it [00:10, 495.09it/s]

4978it [00:11, 495.79it/s]

5000it [00:11, 451.61it/s]

MAE: 3.79

LSTM Regression Model¶

In [6]:

class LstmModule(nn.Module):

    def __init__(self, n_features, hidden_size=1):
        super().__init__()
        self.n_features = n_features
        self.hidden_size = hidden_size
        self.lstm = nn.LSTM(
            input_size=n_features,
            hidden_size=hidden_size,
            num_layers=1,
            bidirectional=False,
        )
        self.fc = nn.Linear(in_features=hidden_size, out_features=1)

    def forward(self, X, **kwargs):
        output, (hn, cn) = self.lstm(
            X
        )  # lstm with input, hidden, and internal state
        return self.fc(output[-1, :])

class LstmModule(nn.Module): def __init__(self, n_features, hidden_size=1): super().__init__() self.n_features = n_features self.hidden_size = hidden_size self.lstm = nn.LSTM( input_size=n_features, hidden_size=hidden_size, num_layers=1, bidirectional=False, ) self.fc = nn.Linear(in_features=hidden_size, out_features=1) def forward(self, X, **kwargs): output, (hn, cn) = self.lstm( X ) # lstm with input, hidden, and internal state return self.fc(output[-1, :])

In [7]:

dataset = datasets.Bikes()
metric = metrics.MAE()

model_pipeline = compose.Select(
    "clouds", "humidity", "pressure", "temperature", "wind"
)
model_pipeline += get_hour | feature_extraction.TargetAgg(
    by=["station", "hour"], how=stats.Mean()
)
model_pipeline |= preprocessing.StandardScaler()
model_pipeline |= RollingRegressor(
    module=LstmModule,
    loss_fn="mse",
    optimizer_fn="sgd",
    window_size=20,
    lr=1e-2,
    hidden_size=32,  # parameters of MyModule can be overwritten
    append_predict=True,
)
model_pipeline

dataset = datasets.Bikes() metric = metrics.MAE() model_pipeline = compose.Select( "clouds", "humidity", "pressure", "temperature", "wind" ) model_pipeline += get_hour | feature_extraction.TargetAgg( by=["station", "hour"], how=stats.Mean() ) model_pipeline |= preprocessing.StandardScaler() model_pipeline |= RollingRegressor( module=LstmModule, loss_fn="mse", optimizer_fn="sgd", window_size=20, lr=1e-2, hidden_size=32, # parameters of MyModule can be overwritten append_predict=True, ) model_pipeline

Out[7]:

['clouds', [...]

Select ( clouds humidity pressure temperature wind )

get_hour

def get_hour(x): x["hour"] = x["moment"].hour return x

y_mean_by_station_and_hour

TargetAgg ( by=['station', 'hour'] how=Mean () target_name="y" )

StandardScaler

StandardScaler ( with_std=True )

RollingRegressor

RollingRegressor ( module=None loss_fn="mse" optimizer_fn="sgd" lr=0.01 is_feature_incremental=False window_size=20 append_predict=True device="cpu" seed=42 )

In [8]:

for x, y in tqdm(dataset.take(5000)):
    y_pred = model_pipeline.predict_one(x)
    metric.update(y_true=y, y_pred=y_pred)
    model_pipeline.learn_one(x=x, y=y)
print(f"MAE: {metric.get():.2f}")

for x, y in tqdm(dataset.take(5000)): y_pred = model_pipeline.predict_one(x) metric.update(y_true=y, y_pred=y_pred) model_pipeline.learn_one(x=x, y=y) print(f"MAE: {metric.get():.2f}")

0it [00:00, ?it/s]

43it [00:00, 424.32it/s]

93it [00:00, 467.02it/s]

145it [00:00, 488.83it/s]

194it [00:00, 484.04it/s]

248it [00:00, 500.65it/s]

299it [00:00, 503.61it/s]

350it [00:00, 479.84it/s]

400it [00:00, 485.69it/s]

449it [00:00, 479.78it/s]

499it [00:01, 483.66it/s]

550it [00:01, 489.44it/s]

601it [00:01, 495.08it/s]

651it [00:01, 496.08it/s]

701it [00:01, 492.19it/s]

751it [00:01, 493.28it/s]

801it [00:01, 494.40it/s]

852it [00:01, 497.34it/s]

902it [00:01, 489.25it/s]

951it [00:01, 487.44it/s]

1001it [00:02, 490.96it/s]

1051it [00:02, 492.80it/s]

1101it [00:02, 493.00it/s]

1151it [00:02, 480.43it/s]

1201it [00:02, 485.42it/s]

1252it [00:02, 489.93it/s]

1302it [00:02, 491.20it/s]

1352it [00:02, 483.61it/s]

1401it [00:02, 485.37it/s]

1450it [00:02, 485.49it/s]

1500it [00:03, 487.10it/s]

1549it [00:03, 486.61it/s]

1598it [00:03, 484.81it/s]

1647it [00:03, 475.10it/s]

1697it [00:03, 481.17it/s]

1747it [00:03, 483.96it/s]

1797it [00:03, 487.37it/s]

1846it [00:03, 486.54it/s]

1895it [00:03, 479.73it/s]

1944it [00:03, 479.98it/s]

1994it [00:04, 484.49it/s]

2044it [00:04, 488.27it/s]

2093it [00:04, 487.41it/s]

2142it [00:04, 477.22it/s]

2192it [00:04, 481.26it/s]

2241it [00:04, 482.43it/s]

2290it [00:04, 449.76it/s]

2336it [00:04, 441.63it/s]

2381it [00:04, 439.74it/s]

2429it [00:05, 449.24it/s]

2479it [00:05, 462.03it/s]

2528it [00:05, 469.25it/s]

2576it [00:05, 471.22it/s]

2624it [00:05, 471.51it/s]

2673it [00:05, 473.39it/s]

2721it [00:05, 456.79it/s]

2770it [00:05, 465.84it/s]

2817it [00:05, 464.86it/s]

2865it [00:05, 468.90it/s]

2917it [00:06, 482.63it/s]

2968it [00:06, 490.03it/s]

3018it [00:06, 491.49it/s]

3069it [00:06, 496.58it/s]

3120it [00:06, 497.89it/s]

3170it [00:06, 498.06it/s]

3222it [00:06, 502.33it/s]

3273it [00:06, 495.73it/s]

3324it [00:06, 499.53it/s]

3376it [00:06, 503.68it/s]

3427it [00:07, 505.24it/s]

3479it [00:07, 508.46it/s]

3530it [00:07, 502.86it/s]

3581it [00:07, 502.88it/s]

3632it [00:07, 503.35it/s]

3683it [00:07, 503.22it/s]

3735it [00:07, 506.44it/s]

3786it [00:07, 495.98it/s]

3836it [00:07, 495.69it/s]

3886it [00:08, 494.41it/s]

3937it [00:08, 498.61it/s]

3988it [00:08, 501.01it/s]

4039it [00:08, 497.15it/s]

4089it [00:08, 493.27it/s]

4141it [00:08, 500.15it/s]

4192it [00:08, 501.72it/s]

4244it [00:08, 505.43it/s]

4295it [00:08, 499.87it/s]

4346it [00:08, 499.81it/s]

4398it [00:09, 503.11it/s]

4449it [00:09, 496.05it/s]

4501it [00:09, 500.42it/s]

4552it [00:09, 493.78it/s]

4602it [00:09, 490.95it/s]

4654it [00:09, 498.03it/s]

4704it [00:09, 498.05it/s]

4755it [00:09, 498.76it/s]

4805it [00:09, 497.63it/s]

4855it [00:09, 492.67it/s]

4905it [00:10, 490.55it/s]

4955it [00:10, 490.07it/s]

5000it [00:10, 488.04it/s]

MAE: 2.78