Applying Transformer methods for predicting energy consumption. Founded by https://navgreen.gr.
To get the necessary libraries refer to the poetry configuration provided.
Data is publicly available at: http://cogsys.imm.dtu.dk/propulsionmodelling/data.html.
We aggregate the available data to 3 minutes, maintaining their mean and standard deviation.
The data processing is available at ./data_creation and the processed data at ./data_creation/data
-
./base/hist_data_analysis/transformer/loader.py : Provides the data loading for the inference and training. It processes the timeseries with a fixed length, adds the time representation, assures the correct ordering of the timeseries and handles the training, validation and testing data splits.
-
./base/hist_data_analysis/transformer/model.py : Provides the simple attention-based, Transformer model for the regression task.
-
./base/hist_data_analysis/transformer/train_eval.py : Provides the training and evaluation pipeline. It, also, stores the training information and evaluation results and provides some basic visualization relevant to the task.
To run the pipeline, one simply can execute this script.
-
./base/hist_data_analysis/transformer/eval.py : Provides a more formal evaluation for the regression task.
-
./base/hist_data_analysis/transformer/utils.py : Provides utility functions for the loading, storing and plotting as well as different loss function implementations.