- ./data contains .tif for elevation data. curl and extract data from 1982-2010 into ./data/old and 2012-2020 into ./data/new
- ./deprecated contains Rscripts we no longer use, but we're keeping as backup. Never delete!
- ./images contains charts and graphs from tests/experiments
- climate_norm_experiment tests climate norm versus quantile matching performance on all US pixels
- CNN_Dev
- CNN Practice
- data_sampling_dev
- EDA_state exploratory data analysis for some other US states
- EDA_Vermont exploratory data analysis for Vermont
- LSTM_dev
- make_data_dev
- make_data_slice
- make_data_with_state is similar to make_train_data, but you can filter cells by state (rather than random sampling).
- make_test_data is similar to make_train, but requires a train data slice, and generates a similar slice from the test data.
- make_test_data_CNN will generate a 10 x 10 pixel testing subset of US data (includes Vermont)
- make_train_data will clean, subset, and store a data slice from the old data. No fancy formatting, you'll have to dig into the code...
- made_train_data_CNN will generate a 10 x 10 pixel training subset of US data (includes Vermont)
- ML contains ACF and LSTM resources.
- PCA conducts a principal components analysis and visualizes results for training + testing data
- process_data_CNN shapes train and test data to be the correct input shape for CNN
- results_visualizations visualizes model results
- rf_testing_multi_cell will take a dataframe of RFs as input and train and test datasets, then do MSE analysis on the test data against GT, QM, and climate norm.
- rf_training_multi_cell can build and tune RFs on multiple points simultaneously. This is the alternate to:
- rf_training_single_cell, which builds and tunes a separate RF for every cell in the train data.
- state-bias determines bias on the full old dataset by state using random sampling.
- stats_testing has some experimental statistical methods such as GAM.