How to Extract Training and Prediction Matrices¶
While the primary goal of building forecasting models is to predict future values, it is equally important to evaluate if the model is effectively learning from the training data. Analyzing predictions on the training data or exploring the prediction matrices is crucial for assessing model performance and understanding areas for optimization. This process can help identify issues like overfitting or underfitting, as well as provide deeper insights into the model’s decision-making process.
Training matrices
Training matrices contain the input features used by the model during the training process. These matrices are essential for understanding how the model interprets patterns and relationships within the data. They typically include the lagged variables, window features and exogenous variables. By extracting and analyzing these matrices, you can ensure that the input data is correctly structured and aligned with the model’s requirements.
Prediction matrices
Prediction matrices are used to generate forecasts for future values. These matrices incorporate the features necessary for making predictions, such as recent observations (lags), window features and any exogenous variables. Understanding the structure of these matrices is important for debugging and for validating the model’s future predictions.
⚠ Warning
If any data transformations and/or differentiation, are applied, they will affect the output matrices. Consequently, the predictions generated in this transformed scale may require additional steps to revert back to the original data scale.
Libraries and data¶
# Libraries
# ==============================================================================
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import Ridge
from lightgbm import LGBMRegressor
from skforecast.datasets import fetch_dataset
from skforecast.preprocessing import RollingFeatures
from skforecast.recursive import ForecasterRecursive
from skforecast.direct import ForecasterDirect
from skforecast.recursive import ForecasterRecursiveMultiSeries
from skforecast.direct import ForecasterDirectMultiVariate
# Download data single series
# ==============================================================================
data = fetch_dataset(
name="h2o", kwargs_read_csv={"names": ["y", "datetime"], "header": 0}
)
print("")
data['datetime'] = pd.to_datetime(data['datetime'], format='%Y-%m-%d')
data = data.set_index('datetime')
data = data.asfreq('MS')
data = data.sort_index()
# Download data ForecasterRecursiveMultiSeries
# ==============================================================================
data_multiseries = fetch_dataset(name="items_sales")
print("")
# Download data ForecasterDirectMultiVariate
# ==============================================================================
data_multivariate = fetch_dataset(name="air_quality_valencia_no_missing")
h2o --- Monthly expenditure ($AUD) on corticosteroid drugs that the Australian health system had between 1991 and 2008. Hyndman R (2023). fpp3: Data for Forecasting: Principles and Practice(3rd Edition). http://pkg.robjhyndman.com/fpp3package/,https://github.com/robjhyndman /fpp3package, http://OTexts.com/fpp3. Shape of the dataset: (204, 2) items_sales ----------- Simulated time series for the sales of 3 different items. Simulated data. Shape of the dataset: (1097, 3) air_quality_valencia_no_missing ------------------------------- Hourly measures of several air chemical pollutant (pm2.5, co, no, no2, pm10, nox, o3, veloc. (air speed), direc. (air direction), so2) at Valencia city. Units are (µg/m3) for pm2.5, no, no2, pm10, so2, (mg/m3) for co, (m/s) for veloc. and (degrees) for direc. Missing values have been removed using linear interpolation. Red de Vigilancia y Control de la Contaminación Atmosférica, 46250054-València - Centre, https://mediambient.gva.es/es/web/calidad-ambiental/datos-historicos. Shape of the dataset: (26304, 10)
ForecasterRecursive¶
# Data
# ==============================================================================
display(data.head(3))
# Plot
# ==============================================================================
fig, ax = plt.subplots(figsize=(6, 3))
data['y'].plot(ax=ax)
ax.legend()
plt.show()
y | |
---|---|
datetime | |
1991-07-01 | 0.429795 |
1991-08-01 | 0.400906 |
1991-09-01 | 0.432159 |
# Create and fit forecaster
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterRecursive(
regressor = LGBMRegressor(random_state=123, verbose=-1),
lags = 5,
window_features = window_features
)
forecaster.fit(y=data['y'])
forecaster
ForecasterRecursive
General Information
- Regressor: LGBMRegressor
- Lags: [1 2 3 4 5]
- Window features: ['roll_mean_5', 'roll_sum_5']
- Window size: 5
- Exogenous included: False
- Weight function included: False
- Differentiation order: None
- Creation date: 2024-11-08 16:14:10
- Last fit date: 2024-11-08 16:14:10
- Skforecast version: 0.14.0
- Python version: 3.11.10
- Forecaster id: None
Exogenous Variables
-
None
Data Transformations
- Transformer for y: None
- Transformer for exog: None
Training Information
- Training range: [Timestamp('1991-07-01 00:00:00'), Timestamp('2008-06-01 00:00:00')]
- Training index type: DatetimeIndex
- Training index frequency: MS
Regressor Parameters
-
{'boosting_type': 'gbdt', 'class_weight': None, 'colsample_bytree': 1.0, 'importance_type': 'split', 'learning_rate': 0.1, 'max_depth': -1, 'min_child_samples': 20, 'min_child_weight': 0.001, 'min_split_gain': 0.0, 'n_estimators': 100, 'n_jobs': None, 'num_leaves': 31, 'objective': None, 'random_state': 123, 'reg_alpha': 0.0, 'reg_lambda': 0.0, 'subsample': 1.0, 'subsample_for_bin': 200000, 'subsample_freq': 0, 'verbose': -1}
Fit Kwargs
-
{}
# Create training matrices
# ==============================================================================
X_train, y_train = forecaster.create_train_X_y(y=data['y'])
# Predictors matrix
# ==============================================================================
X_train.head(3)
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | |
---|---|---|---|---|---|---|---|
datetime | |||||||
1991-12-01 | 0.502369 | 0.492543 | 0.432159 | 0.400906 | 0.429795 | 0.451554 | 2.257772 |
1992-01-01 | 0.602652 | 0.502369 | 0.492543 | 0.432159 | 0.400906 | 0.486126 | 2.430629 |
1992-02-01 | 0.660119 | 0.602652 | 0.502369 | 0.492543 | 0.432159 | 0.537968 | 2.689842 |
# Target variable matrix
# ==============================================================================
y_train.head(3)
datetime 1991-12-01 0.602652 1992-01-01 0.660119 1992-02-01 0.336220 Freq: MS, Name: y, dtype: float64
We can obtain the training predictions using the predict
method of the regressor stored inside the forecaster object. By examining the predictions on the training data, analysts can get a better understanding of how the model is performing and make adjustments as necessary.
# Training predictions using the internal regressor
# ==============================================================================
predictions_training = forecaster.regressor.predict(X_train)
predictions_training[:4]
array([0.49322601, 0.6376049 , 0.58531495, 0.44962278])
Skforecast provides the create_predict_X
method to generate the matrices that the forecaster is using to make predictions. This method can be used to gain insight into the specific data manipulations that occur during the prediction process.
# Create input matrix for predict method
# ==============================================================================
X_predict = forecaster.create_predict_X(steps=5)
X_predict
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | |
---|---|---|---|---|---|---|---|
2008-07-01 | 0.762137 | 0.816255 | 0.827887 | 0.649435 | 0.761822 | 0.763507 | 3.817536 |
2008-08-01 | 0.865361 | 0.762137 | 0.816255 | 0.827887 | 0.649435 | 0.784215 | 3.921075 |
2008-09-01 | 0.878167 | 0.865361 | 0.762137 | 0.816255 | 0.827887 | 0.829961 | 4.149806 |
2008-10-01 | 0.806708 | 0.878167 | 0.865361 | 0.762137 | 0.816255 | 0.825726 | 4.128628 |
2008-11-01 | 0.873597 | 0.806708 | 0.878167 | 0.865361 | 0.762137 | 0.837194 | 4.185970 |
# Predict using the internal regressor
# ==============================================================================
predictions = forecaster.regressor.predict(X_predict)
predictions
array([0.86536052, 0.87816664, 0.80670845, 0.87359717, 0.96601636])
ForecasterDirect¶
# Create and fit forecaster
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterDirect(
regressor = Ridge(random_state=123),
steps = 3,
lags = 5,
window_features = window_features
)
forecaster.fit(y=data['y'])
Two steps are required to extract the training matrices. One to create the entire training matrix and a second to subset the data needed for each model (step).
# Create the whole train matrix
# ==============================================================================
X_train, y_train = forecaster.create_train_X_y(y=data['y'])
# Extract X and y for step 1
X_train_1, y_train_1 = forecaster.filter_train_X_y_for_step(
step = 1,
X_train = X_train,
y_train = y_train,
remove_suffix = False
)
X_train_1.head(4)
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | |
---|---|---|---|---|---|---|---|
datetime | |||||||
1991-12-01 | 0.502369 | 0.492543 | 0.432159 | 0.400906 | 0.429795 | 0.451554 | 2.257772 |
1992-01-01 | 0.602652 | 0.502369 | 0.492543 | 0.432159 | 0.400906 | 0.486126 | 2.430629 |
1992-02-01 | 0.660119 | 0.602652 | 0.502369 | 0.492543 | 0.432159 | 0.537968 | 2.689842 |
1992-03-01 | 0.336220 | 0.660119 | 0.602652 | 0.502369 | 0.492543 | 0.518781 | 2.593903 |
# Target variable matrix for step 1
# ==============================================================================
y_train_1.head(3)
datetime 1991-12-01 0.602652 1992-01-01 0.660119 1992-02-01 0.336220 Freq: MS, Name: y_step_1, dtype: float64
# Internal regressors {step: regressor}
# ==============================================================================
forecaster.regressors_
{1: Ridge(random_state=123), 2: Ridge(random_state=123), 3: Ridge(random_state=123)}
# Step 1 training predictions using the internal regressor
# ==============================================================================
predictions_training = forecaster.regressors_[1].predict(X_train_1)
predictions_training[:4]
array([0.5960254 , 0.6592509 , 0.70209408, 0.50312286])
# Create input matrix for predict method
# ==============================================================================
X_predict = forecaster.create_predict_X(steps=None) # All steps
X_predict
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | |
---|---|---|---|---|---|---|---|
2008-07-01 | 0.762137 | 0.816255 | 0.827887 | 0.649435 | 0.761822 | 0.763507 | 3.817536 |
2008-08-01 | 0.762137 | 0.816255 | 0.827887 | 0.649435 | 0.761822 | 0.763507 | 3.817536 |
2008-09-01 | 0.762137 | 0.816255 | 0.827887 | 0.649435 | 0.761822 | 0.763507 | 3.817536 |
# Step 1 predictions using the internal regressor
# ==============================================================================
predictions = forecaster.regressors_[1].predict(X_predict)
predictions
array([0.78198225, 0.78198225, 0.78198225])
Creating matrices when including transformations¶
If any data transformations and/or differentiation, are applied, they will affect the output matrices. Consequently, the predictions generated in this transformed scale may require additional steps to revert back to the original data scale.
# Create and fit ForecasterRecursive
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterRecursive(
regressor = LGBMRegressor(random_state=123, verbose=-1),
lags = 5,
window_features = window_features,
transformer_y = StandardScaler(),
differentiation = 1
)
forecaster.fit(y=data['y'])
# Training predictions with transformations
# ==============================================================================
X_train_transformed, y_train_transformed = forecaster.create_train_X_y(y=data['y'])
# Training predictions using the internal regressor
predictions_transformed = forecaster.regressor.predict(X_train_transformed)
# Revert differentiation (only if differentiation is not None)
predictions_transformed = forecaster.differentiator.inverse_transform_training(predictions_transformed)
# Revert transformation (only if transformer_y is not None)
predictions_training = forecaster.transformer_y.inverse_transform(predictions_transformed.reshape(-1, 1))
predictions_training.ravel()[:4]
array([0.5547262 , 0.3597327 , 0.39960716, 0.42227145])
# Predict using the internal regressor with transformation
# ==============================================================================
X_predict_transformed = forecaster.create_predict_X(steps=5)
# Predict using the internal regressor
predictions_transformed = forecaster.regressor.predict(X_predict_transformed)
# Revert differentiation (only if differentiation is not None)
predictions_transformed = forecaster.differentiator.inverse_transform_next_window(predictions_transformed)
# Revert transformation (only if transformer_y is not None)
predictions = forecaster.transformer_y.inverse_transform(predictions_transformed.reshape(-1, 1))
predictions.ravel()[:4]
c:\Users\jaesc2\Miniconda3\envs\skforecast_py11_2\Lib\site-packages\skforecast\recursive\_forecaster_recursive.py:1385: DataTransformationWarning: The output matrix is in the transformed scale due to the inclusion of transformations or differentiation in the Forecaster. As a result, any predictions generated using this matrix will also be in the transformed scale. Please refer to the documentation for more details: https://skforecast.org/latest/user_guides/training-and-prediction-matrices.html You can suppress this warning using: warnings.simplefilter('ignore', category=DataTransformationWarning) warnings.warn(
array([0.88563047, 0.62235217, 0.54433454, 0.56906843])
As before, when using a ForecasterDirect
, two steps are required to extract the training matrices. One to create the entire training matrix and a second to subset the data needed for each model (step).
⚠ Warning
If the ForecasterDirect
includes differentiation, the model in step 1 must be used if you want to reverse the differentiation of the training time series with the inverse_transform_training
method.
# Create and fit ForecasterDirect
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterDirect(
regressor = Ridge(random_state=123),
steps = 3,
lags = 5,
window_features = window_features,
transformer_y = StandardScaler(),
differentiation = 1
)
forecaster.fit(y=data['y'])
# Training predictions with transformations
# ==============================================================================
X_train_transformed, y_train_transformed = forecaster.create_train_X_y(y=data['y'])
# Extract X and y for step 1
X_train_transformed_1, y_train_transformed_1 = forecaster.filter_train_X_y_for_step(
step = 1,
X_train = X_train_transformed,
y_train = y_train_transformed,
remove_suffix = False
)
# Training predictions using the internal regressor for step 1
predictions_transformed = forecaster.regressors_[1].predict(X_train_transformed_1)
# Revert differentiation (only if differentiation is not None)
predictions_transformed = forecaster.differentiator.inverse_transform_training(predictions_transformed)
# Revert transformation (only if transformer_y is not None)
predictions_training = forecaster.transformer_y.inverse_transform(predictions_transformed.reshape(-1, 1))
predictions_training.ravel()[:4]
array([0.58659215, 0.55767068, 0.58243553, 0.58782361])
# Predict using the internal regressor with transformation
# ==============================================================================
X_predict_transformed = forecaster.create_predict_X(steps=None) # All steps
# Predict using the internal regressor for step 1
predictions_transformed = forecaster.regressors_[1].predict(X_predict_transformed)
# Revert differentiation (only if differentiation is not None)
predictions_transformed = forecaster.differentiator.inverse_transform_next_window(predictions_transformed)
# Revert transformation (only if transformer_y is not None)
predictions = forecaster.transformer_y.inverse_transform(predictions_transformed.reshape(-1, 1))
predictions.ravel()[:4]
c:\Users\jaesc2\Miniconda3\envs\skforecast_py11_2\Lib\site-packages\skforecast\direct\_forecaster_direct.py:1388: DataTransformationWarning: The output matrix is in the transformed scale due to the inclusion of transformations or differentiation in the Forecaster. As a result, any predictions generated using this matrix will also be in the transformed scale. Please refer to the documentation for more details: https://skforecast.org/latest/user_guides/training-and-prediction-matrices.html You can suppress this warning using: warnings.simplefilter('ignore', category=DataTransformationWarning) warnings.warn(
array([0.85739057, 0.95264414, 1.04789772])
💡 Tip
To reverse the data transformation, you can also use one of these skforecast functions: transform_numpy
, transform_series
, transform_dataframe
.
from skforecast.utils import transform_numpy
predictions = transform_numpy(
array = predictions_transformed,
transformer = forecaster.transformer_y,
fit = False,
inverse_transform = True
)
ForecasterRecursiveMultiSeries¶
# Data
# ==============================================================================
display(data_multiseries.head(3))
# Plot
# ==============================================================================
fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(9, 4), sharex=True)
for i, col in enumerate(data_multiseries.columns):
data_multiseries[col].plot(ax=axes[i])
axes[i].set_xlabel('')
axes[i].set_ylabel('sales')
axes[i].set_title(col)
fig.tight_layout()
plt.show();
item_1 | item_2 | item_3 | |
---|---|---|---|
date | |||
2012-01-01 | 8.253175 | 21.047727 | 19.429739 |
2012-01-02 | 22.777826 | 26.578125 | 28.009863 |
2012-01-03 | 27.549099 | 31.751042 | 32.078922 |
# Create and fit forecaster
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterRecursiveMultiSeries(
regressor = LGBMRegressor(random_state=123, verbose=-1),
lags = 5,
window_features = window_features,
encoding = 'ordinal'
)
forecaster.fit(series=data_multiseries)
forecaster
ForecasterRecursiveMultiSeries
General Information
- Regressor: LGBMRegressor
- Lags: [1 2 3 4 5]
- Window features: ['roll_mean_5', 'roll_sum_5']
- Window size: 5
- Series encoding: ordinal
- Exogenous included: False
- Weight function included: False
- Series weights: None
- Differentiation order: None
- Creation date: 2024-11-08 16:14:11
- Last fit date: 2024-11-08 16:14:11
- Skforecast version: 0.14.0
- Python version: 3.11.10
- Forecaster id: None
Exogenous Variables
-
None
Data Transformations
- Transformer for series: None
- Transformer for exog: None
Training Information
- Series names (levels): item_1, item_2, item_3
- Training range: 'item_1': ['2012-01-01', '2015-01-01'], 'item_2': ['2012-01-01', '2015-01-01'], 'item_3': ['2012-01-01', '2015-01-01']
- Training index type: DatetimeIndex
- Training index frequency: D
Regressor Parameters
-
{'boosting_type': 'gbdt', 'class_weight': None, 'colsample_bytree': 1.0, 'importance_type': 'split', 'learning_rate': 0.1, 'max_depth': -1, 'min_child_samples': 20, 'min_child_weight': 0.001, 'min_split_gain': 0.0, 'n_estimators': 100, 'n_jobs': None, 'num_leaves': 31, 'objective': None, 'random_state': 123, 'reg_alpha': 0.0, 'reg_lambda': 0.0, 'subsample': 1.0, 'subsample_for_bin': 200000, 'subsample_freq': 0, 'verbose': -1}
Fit Kwargs
-
{}
# Create training matrices
# ==============================================================================
X_train, y_train = forecaster.create_train_X_y(series=data_multiseries)
Depending on the series encoding selected, the column(s) generated to identify the series to which the observations belong may be different. In this case, the column _level_skforecast
is generated as encoding = 'ordinal'
.
# Predictors matrix
# ==============================================================================
X_train.head(3)
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | _level_skforecast | |
---|---|---|---|---|---|---|---|---|
date | ||||||||
2012-01-06 | 21.379238 | 25.895533 | 27.549099 | 22.777826 | 8.253175 | 21.170974 | 105.854870 | 0 |
2012-01-07 | 21.106643 | 21.379238 | 25.895533 | 27.549099 | 22.777826 | 23.741668 | 118.708338 | 0 |
2012-01-08 | 20.533871 | 21.106643 | 21.379238 | 25.895533 | 27.549099 | 23.292877 | 116.464384 | 0 |
# Target variable matrix
# ==============================================================================
y_train.head(3)
date 2012-01-06 21.106643 2012-01-07 20.533871 2012-01-08 20.069327 Name: y, dtype: float64
We can obtain the training predictions using the predict
method of the regressor stored inside the forecaster object. By examining the predictions on the training data, analysts can get a better understanding of how the model is performing and make adjustments as necessary.
# Training predictions using the internal regressor
# ==============================================================================
predictions_training = forecaster.regressor.predict(X_train)
predictions_training[:4]
array([19.54628549, 22.29989602, 20.10135048, 20.97563208])
Skforecast provides the create_predict_X
method to generate the matrices that the forecaster is using to make predictions. This method can be used to gain insight into the specific data manipulations that occur during the prediction process.
# Create input matrix for predict method
# ==============================================================================
X_predict_dict = forecaster.create_predict_X(steps=5, levels=None) # All levels
# Check 'item_1' matrix
X_predict_item_1 = X_predict_dict['item_1']
X_predict_item_1.head()
lag_1 | lag_2 | lag_3 | lag_4 | lag_5 | roll_mean_5 | roll_sum_5 | _level_skforecast | |
---|---|---|---|---|---|---|---|---|
2015-01-02 | 10.496302 | 18.721223 | 18.857026 | 19.611623 | 17.329233 | 17.003081 | 85.015406 | 0.0 |
2015-01-03 | 13.614696 | 10.496302 | 18.721223 | 18.857026 | 19.611623 | 16.260174 | 81.300869 | 0.0 |
2015-01-04 | 14.526244 | 13.614696 | 10.496302 | 18.721223 | 18.857026 | 15.243098 | 76.215490 | 0.0 |
2015-01-05 | 16.802037 | 14.526244 | 13.614696 | 10.496302 | 18.721223 | 14.832100 | 74.160501 | 0.0 |
2015-01-06 | 13.888023 | 16.802037 | 14.526244 | 13.614696 | 10.496302 | 13.865460 | 69.327302 | 0.0 |
# Predict 'item_1' using the internal regressor
# ==============================================================================
predictions_item_1 = forecaster.regressor.predict(X_predict_item_1)
predictions_item_1
array([13.61469596, 14.5262436 , 16.80203691, 13.88802319, 15.13547167])
ForecasterDirectMultiVariate¶
# Data
# ==============================================================================
display(data_multivariate.head(3))
# Plot
# ==============================================================================
fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(9, 4), sharex=True)
for i, col in enumerate(data_multivariate.columns[:3]):
data_multivariate[col].plot(ax=axes[i])
axes[i].set_xlabel('')
axes[i].set_ylabel('sales')
axes[i].set_title(col)
fig.tight_layout()
plt.show();
pm2.5 | co | no | no2 | pm10 | nox | o3 | veloc. | direc. | so2 | |
---|---|---|---|---|---|---|---|---|---|---|
datetime | ||||||||||
2019-01-01 00:00:00 | 19.0 | 0.2 | 3.0 | 36.0 | 22.0 | 40.0 | 16.0 | 0.5 | 262.0 | 8.0 |
2019-01-01 01:00:00 | 26.0 | 0.1 | 2.0 | 40.0 | 32.0 | 44.0 | 6.0 | 0.6 | 248.0 | 8.0 |
2019-01-01 02:00:00 | 31.0 | 0.1 | 11.0 | 42.0 | 36.0 | 58.0 | 3.0 | 0.3 | 224.0 | 8.0 |
# Create and fit forecaster
# ==============================================================================
window_features = RollingFeatures(
stats = ['mean', 'sum'],
window_sizes = [5, 5]
)
forecaster = ForecasterDirectMultiVariate(
regressor = Ridge(random_state=123),
level = 'co',
steps = 3,
lags = 3,
window_features = window_features
)
forecaster.fit(series=data_multivariate)
forecaster
ForecasterDirectMultiVariate
General Information
- Regressor: Ridge
- Target series (level): co
- Lags: [1 2 3]
- Window features: ['roll_mean_5', 'roll_sum_5']
- Window size: 5
- Maximum steps to predict: 3
- Exogenous included: False
- Weight function included: False
- Differentiation order: None
- Creation date: 2024-11-08 16:14:12
- Last fit date: 2024-11-08 16:14:12
- Skforecast version: 0.14.0
- Python version: 3.11.10
- Forecaster id: None
Exogenous Variables
-
None
Data Transformations
- Transformer for series: StandardScaler()
- Transformer for exog: None
Training Information
- Target series (level): co
- Multivariate series: pm2.5, co, no, no2, pm10, nox, o3, veloc., direc., so2
- Training range: [Timestamp('2019-01-01 00:00:00'), Timestamp('2021-12-31 23:00:00')]
- Training index type: DatetimeIndex
- Training index frequency: h
Regressor Parameters
-
{'alpha': 1.0, 'copy_X': True, 'fit_intercept': True, 'max_iter': None, 'positive': False, 'random_state': 123, 'solver': 'auto', 'tol': 0.0001}
Fit Kwargs
-
{}
# Create the whole train matrix
# ==============================================================================
X_train, y_train = forecaster.create_train_X_y(series=data_multivariate)
# Extract X and y for step 1
X_train_1, y_train_1 = forecaster.filter_train_X_y_for_step(
step = 1,
X_train = X_train,
y_train = y_train,
remove_suffix = False
)
print("Columns :", list(X_train_1.columns))
X_train_1.head(3)
Columns : ['pm2.5_lag_1', 'pm2.5_lag_2', 'pm2.5_lag_3', 'pm2.5_roll_mean_5', 'pm2.5_roll_sum_5', 'co_lag_1', 'co_lag_2', 'co_lag_3', 'co_roll_mean_5', 'co_roll_sum_5', 'no_lag_1', 'no_lag_2', 'no_lag_3', 'no_roll_mean_5', 'no_roll_sum_5', 'no2_lag_1', 'no2_lag_2', 'no2_lag_3', 'no2_roll_mean_5', 'no2_roll_sum_5', 'pm10_lag_1', 'pm10_lag_2', 'pm10_lag_3', 'pm10_roll_mean_5', 'pm10_roll_sum_5', 'nox_lag_1', 'nox_lag_2', 'nox_lag_3', 'nox_roll_mean_5', 'nox_roll_sum_5', 'o3_lag_1', 'o3_lag_2', 'o3_lag_3', 'o3_roll_mean_5', 'o3_roll_sum_5', 'veloc._lag_1', 'veloc._lag_2', 'veloc._lag_3', 'veloc._roll_mean_5', 'veloc._roll_sum_5', 'direc._lag_1', 'direc._lag_2', 'direc._lag_3', 'direc._roll_mean_5', 'direc._roll_sum_5', 'so2_lag_1', 'so2_lag_2', 'so2_lag_3', 'so2_roll_mean_5', 'so2_roll_sum_5']