business-science / pymodeltime

Pymodeltime offers a unified framework tailored to address a broad spectrum of requirements, including time series forecasting and various machine learning models.

pymodeltime

• Univariate Regression - Open in Colab
• Multiple_regression - Open in Colab
• Demand Forecasting - Open in Colab
• ml_forecast_pymodeltime - Open in Colab

Installation

!pip install git+https://github.com/business-science/pymodeltime.git

Usage Examples

Importing Necessary Modules

from pymodeltime import ArimaReg, ProphetReg
from pymodeltime import ModelTimeTable, ModelTimeAccuracy, ModelTimeCalibration, ModelTimeForecast, ModelTimeRefit
from pymodeltime import MLModelWrapper, H2OAutoMLWrapper
from sklearn.ensemble import RandomForestRegressor
from xgboost import XGBRegressor

Create instances of the ML models

rf_model = RandomForestRegressor()
ml_rf_wrapper = MLModelWrapper(rf_model, feature_columns, "Random Forest")
##Machine Learning Model Integration
xgb_model = XGBRegressor()
ml_xgb_wrapper = MLModelWrapper(xgb_model, feature_columns, "XGBoost")

# Fit the models
ml_rf_wrapper.fit(train_data, y_train)
ml_xgb_wrapper.fit(train_data, y_train)

AutoGluon models

from pymodeltime import AutoGluonTabularWrapper

auto_gluon_wrapper = AutoGluonTabularWrapper(target_column="GDP")
auto_gluon_wrapper.fit(train_data)
quantile_predictions = auto_gluon_wrapper.predict_quantiles(test_data)

H2O AutoML Integration

h2o.init()

# Convert to H2OFrame and train
h2o_train = h2o.H2OFrame(train_data)
automl = H2OAutoML(max_models=10, seed=1, max_runtime_secs=300)
automl.train(x=feature_columns, y='GDP', training_frame=h2o_train)

# Extract the leaderboard
lb = automl.leaderboard
lb_df = lb.as_data_frame()
print(lb_df.head())

# Get the best model
best_h2o_model = automl.leader
h2o_automl_wrapper = H2OAutoMLWrapper(best_h2o_model, target_column='GDP')

Prophet Model for Multivariate Time Series

prophet_model_multi = ProphetReg(seasonality_yearly=True, seasonality_weekly=True,
                                 seasonality_daily=False, changepoint_range=0.5,
                                 prior_scale_seasonality=5.0, season='multiplicative',
                                 interval_width=0.95)

prophet_model_multi.fit(train_data, target_column='GDP', date_column='date', regressors=feature_columns)

Model Calibration and Accuracy Evaluation

modeltime_table = ModelTimeTable(h2o_automl_wrapper, ml_xgb_wrapper, ml_rf_wrapper, prophet_model_multi)

Model Calibration

model_time_calibrator = ModelTimeCalibration(modeltime_table, test_data, target_column='GDP')
model_time_calibrator.calibrate()
calibration_results_df = model_time_calibrator.get_calibration_results()
print(calibration_results_df)

Model Accuracy

modeltime_accuracy = ModelTimeAccuracy(modeltime_table, test_data, target_column='GDP')
accuracy_results = modeltime_accuracy.calculate_accuracy()
print(accuracy_results)