graph LR
exp_exp_basic_ExpBasic["exp.exp_basic.ExpBasic"]
exp_exp_long_term_forecasting_ExpLongTermForecasting["exp.exp_long_term_forecasting.ExpLongTermForecasting"]
exp_exp_short_term_forecasting_ExpShortTermForecasting["exp.exp_short_term_forecasting.ExpShortTermForecasting"]
exp_exp_anomaly_detection_ExpAnomalyDetection["exp.exp_anomaly_detection.ExpAnomalyDetection"]
exp_exp_classification_ExpClassification["exp.exp_classification.ExpClassification"]
exp_exp_imputation_ExpImputation["exp.exp_imputation.ExpImputation"]
models_TimeMixer["models.TimeMixer"]
exp_exp_basic_ExpBasic -- "orchestrates" --> models_TimeMixer
exp_exp_long_term_forecasting_ExpLongTermForecasting -- "inherits from" --> exp_exp_basic_ExpBasic
exp_exp_short_term_forecasting_ExpShortTermForecasting -- "inherits from" --> exp_exp_basic_ExpBasic
exp_exp_anomaly_detection_ExpAnomalyDetection -- "inherits from" --> exp_exp_basic_ExpBasic
exp_exp_classification_ExpClassification -- "inherits from" --> exp_exp_basic_ExpBasic
exp_exp_imputation_ExpImputation -- "inherits from" --> exp_exp_basic_ExpBasic
exp_exp_long_term_forecasting_ExpLongTermForecasting -- "utilizes" --> models_TimeMixer
exp_exp_short_term_forecasting_ExpShortTermForecasting -- "utilizes" --> models_TimeMixer
exp_exp_anomaly_detection_ExpAnomalyDetection -- "utilizes" --> models_TimeMixer
exp_exp_classification_ExpClassification -- "utilizes" --> models_TimeMixer
exp_exp_imputation_ExpImputation -- "utilizes" --> models_TimeMixer
The Experiment Management Framework subsystem is the core orchestrator for machine learning experiments within the TimeMixer project, managing the entire lifecycle from model building to evaluation across various time series tasks.
The foundational abstract base class for all experiments. It establishes the common interface and core functionalities for experiment setup, model initialization, data loading, training loops, validation, testing, and result saving. It acts as the blueprint for conducting any time series experiment within the TimeMixer framework, ensuring consistency and extensibility.
Related Classes/Methods:
A concrete implementation of ExpBasic specifically designed for long-term time series forecasting experiments. It extends the base class to incorporate task-specific logic, such as appropriate loss functions (e.g., MSE, MAE) and evaluation metrics relevant to long prediction horizons.
Related Classes/Methods:
A concrete implementation of ExpBasic for short-term time series forecasting experiments. Similar to its long-term counterpart, it adapts the general experiment flow for shorter prediction horizons and potentially different evaluation criteria.
Related Classes/Methods:
A concrete implementation of ExpBasic tailored for time series anomaly detection experiments. This component focuses on defining anomaly scoring mechanisms, thresholding strategies, and specific evaluation metrics (e.g., precision, recall, F1-score for anomalies).
Related Classes/Methods:
A concrete implementation of ExpBasic for time series classification experiments. It handles multi-class or binary classification tasks, integrating appropriate loss functions (e.g., cross-entropy) and classification-specific metrics (e.g., accuracy, F1-score).
Related Classes/Methods:
A concrete implementation of ExpBasic for time series imputation experiments. Its primary responsibility is to manage the process of filling missing values in time series data, utilizing metrics like MSE or MAE to evaluate the quality of imputation.
Related Classes/Methods:
The core TimeMixer model, representing the machine learning algorithm that processes time series data. This component is instantiated, trained, validated, and tested by the various Exp* components.
Related Classes/Methods: