This is the code related to the paper "Multi-step probabilistic forecasting model using deep learning parametrized distributions".

Installation

Create a virtualenv in order to not clash with local python libraries, and then install requirements into it:

python3 -m venv p3
. p3/bin/activate
pip install -r requirements.txt

Otherwise, newest version of libraries can be installed using

pip install pandas sklearn tensorflow tensorflow_probability

Running

The script has many options that can be explored through the help option:

python main.py --help

An example run line would be:

python main.py \
    --run_directory test2 \
    --seed 1 \
    --ensemble_number_models 1 \
    --file_name datasets/inf475/substation_load.csv \
    --fillna_method repeat_daily \
    --input_lags 3 \
    --input_series A_DE_CORDOVA__013 APOQUINDO_____013 LA_REINA______013 \
    --input_steps 48 \
    --output_steps 24 \
    --output_series A_DE_CORDOVA__013 LA_REINA______013 \
    --resolution H \
    --resolution_method sum \
    --train_percentage 0.3 \
    --test_size 1000 \
    --validation_percentage 0.1 \
    --number_splits 4 \
    --split_overlap 0.2 \
    --split_position 2 \
    --evaluation_sampler_number 200 \
    --model paperlstmsampled \
    --preprocess minmax \
    --sequential_mini_step -1 \
    --nn_batch_size 128 \
    --nn_dropout_output -1.0 \
    --nn_dropout_recurrence -1.0 \
    --nn_epochs 10000 \
    --nn_l2_regularizer 0.000541369 \
    --nn_learning_rate 9.50585e-05 \
    --nn_optimizer Adam \
    --nn_output_distribution normal \
    --nn_patience 50 \
    --lstm_layers 1 \
    --lstm_nodes 96

Don't hesitate to ask me further help!