README.md

Covid19 predictions through linear regression models

Repo of covid19 prediction Models and Data Visualization.

This work is related to the XPrize Pandemic Response Challenge. The challenge's first phase goal was to develop a model able to predict new cases of Covid19 over 182 countries and over several time horizons. A lot of work regarding the code and the modelling has been made in collaboration with my team 'Transatlantic team' participating in the XPrize Pandemic Response Challenge.

Requirements :

Install pandas, numpy, matplotlib, seaborn, Trendy, scikit-learn.

Scripts:

In both scripts, functions PreProcessing will return the preprocessed dataset. Also, functions GetCovidPredictions will perform the preprocessing, training and prediction according to the specified arguments. It returns both the predictions dataframe and the list of each country's MAE.

BASELINE

Import the COVID_Predictor_baseline class function from script_model_predictions_baseline.py. Call GetCovidPredictions function and specify whether to predict on 4 days, 7 days or 30 days (period), the number of lookback days to take into account nb_lookback_days, whether to train on data up to January 2020 or only up to August 2020 (reduced_train=True) and whether or not to train and predict using different lookback days (test_several_lookbacks, returns only the global MAE).

MODEL WITH ADDITIONAL GOOGLE MOBILITY DATA

Import the COVID_Predictor class function from script_model_predictions_baseline.py by specifying 'RIDGE', otherwise the class's default model is Lasso. Call GetCovidPredictions function and specify whether to predict on 4 days, 7 days or 30 days (period), the number of lookback days to take into account nb_lookback_days, whether to train on data up to January 2020 or only up to August 2020 (reduced_train=True) and whether or not to train and predict using different lookback days (test_several_lookbacks, returns only the global MAE).

VISUALIZATION

Data visualization is presented in the Data_Visualization jupyter notebook. Model results are presented in the jupyter notebooks Model_results and Model_results_reduced_trainingset (where models are trained on data only up to August 2020).