Research project looking into the effectiveness of gas main investments in Poland.

It consists of different machine learning models that calculate the expected prices of different types of gas mains relatively to the most standard model (see config.py) based on the provided training data.

You can see sample results for given data in the results.ipynb notebook. To be able to fully experience the jupyter notebook results view it on the local jupyter notebook or Google Collab.

Data to train the models should consists of data collection year in the first column followed by features specified in the config.py file. All features (and the year) should be in separate columns delimited by the semicolons. Each datapoint should be in separate row. Datapoint collection year is used to group together all of the datapoints from the same year and center them around the standard yearly datapoint to avoid influence of different geopolitical and economic factors (e.g. trade deals or inflation) that are beyond the simple gas mains data.

To configure the training process change appropriate data and models configuration in the config.py file.

To run the training for specified models and specified configuration simply run the train.py script.