Repository for the project that took place during the Hackathon organized by Eleven Strategy (8-12 February 2021). 5 persons have collaborated on this project.
The project regards the prediction of the taxi-out time of aircrafts in a major airport. The taxi-out time corresponds to the time it takes for a plane to leaves a stand and actually take off. A stand is the area where an aircraft is docked and loads the passengers.
Following, you can find a quick documentation on how to use and launch the code, and the main resources we provide for the project.
To preserve privacy and confidentiality, the data is not available. This repository has been made public to showcase the work that has been done.
The main repositories are:
-
main
- The main.py function will call cleaning, joining, preprocessing, training and testing functions.
-
utils
- date2datetime.py: contains two functions that load the airport, weather and aicraft datasets as pandas dataframes and transforms the relevant columns into datetime formats. The first function takes paths as input, while the second one directly takes pandas dataframes.
- date_functions.py: contains 13 functions that were used in exploration (not all of them) for date formatting purposes, or calendar objects extractions. For instance, the function int2datetime() transforms the integer 20210211 into a datetime object corresponding to 11-02-2021.
- geographic_distance.py: contains a single function that computes the distance (in km) between two points, given the respective latitudes and longitudes.
- explainability.py: mostly contains implementations of library 'shap'. Most functions take as input a fitted model and a pandas dataframe of features. The function explainer_xgb() will produce a plot to visualise the impact of features on the target variable. The function get_interation_plot_tree() will produce an interaction plot for tree-based models; get_summary_plot_tree() will produce a summary plot with the impact of features on model output; get_dependence_plot_tree() will produce a partial dependence plot based on a feature selected as input.
- Graph_airport.ipynb: has the construction of a graph with the distances between stands and runways
- add_current_load_airport_N_Q.py: function that computes the N number of aircrafts moving when the airplane starts taxing and also Q, which are the number of aircrafts that finished the movement while the aircraft in focus is still moving.
- add_current_load_runway.py: Returns the dataframe passed as an input where a column representing the load on the runway was added. Here, the current load on the runway accounts for how many planes were taxiing at the same time and heading to the same runway when a specific aircraft starts taxiing.
- add_shortest_path_length.py: Returns the dataframe passed as an input where a column representing the length of the shortest path between a gate and a runway was added.
- dataset_cleaning.py: cleans all the datasets (accepts different modes -> mode='train' or mode='test')
- dataset_joining.py: joins the four main dataframes.
- Preprocessing.py: generates the preprocessing on the merged dataframe. The preprocessing includes the generation of the target variable, the drop of several useless columns, scaling and OneHot Encoding.
- train.py: training function that trains the selected model Light GBM on the training data.
-
resources
- In this folder you would need to add the data with the name specified in the main.py
-
data exploration
- Weather data exploration
- General first exploration
- Airport location exploration
-
models
- In this folder you can save the model trained during the train function