chap-models/Madagascar_ARIMA

A CHAP-compatible implementation of the old ARIMA model from the Madagascar group. # Madagascar_ARIMA This is an implementation of the old ARIMA model for malaria from the Madagascar group. The original code is from gitlab.com and the data is from gitlab.com. The goal is to integrate the ARIMA model they have made, so it can run through CHAP. The framework is a file for training and a seperate file for prediction. They define the functions train_chap and predict_chap and some standard framework underneath for running the model with CHAP. The strictly necessary files are train.R, predict.R, MLproject and requirements.txt. isolated_run.R is only for testing locally. The goal is to have data as the only input and to return samples from the predicted distribuitions for each time point and location. In this implementation I have used $100$ samples for each observation, should be larger in practice. ## Data The dataset already has 43 features. However, we only use three of them to fit the model, as well as indexes for location and time. In data_preperation.R I remove the unneccessary features for this model. Reducing the dataset is not neccessary, but makes it easier to work with. I also rename the features to the naming convention in CHAP, which for instance is "location" instead of "csb"(should be a overview of the naming conventions somewhere). Then I split the data into training data and test data, where we later predict the values for the test data. These are saved as csv files called trainData and futureClimateData respectively. Additionally, CHAP currently also requires a population column so I added a population column filled with ones, this will change when CHAP is updated. ## Training We source some useful helper functions from utils.R, similar functions exist in other R packages. Not necessary to use utils.R. We then mak …