DeepPaSTL: Spatio-Temporal Deep Learning Methods for Predicting Long-Term Pasture Terrains using Synthetic Datasets

Dependencies

• Install and Setup Anaconda

• Clone this repository and cd into root folder of the project.

• Create a new conda environment using the dependency file in the project folder.

conda env create -f environment.yml

• Activate the environment to run the training or testing scripts.

conda activate ag-bay-torch

Arguments

To Run the predictions for the first time, or using a new dataset please ensure that you have the following arguments enabled:

• --load_data: Scales and performs necessary pre-processing of the input data for creating final sliding window inputs for the network.

• --sequence_data: Create sliding windows and stores into a csv file

• --sequence_to_np: Generates numpy arrays and stores in an hdf5 format, for read on access memory, which can be used directly in the Pytorch DataLoader function.

• --in_seq_len VALUE: Input Sequence Length. Expects an int value.

• --out_seq_len VALUE: Output Sequence Length. Expects an int value.

• --device cpu: To run inference or training on CPU, else default is at cuda.

• --exp_name MODEL_NAME: Experiment name to load the model during inference, or save the mode as during training.

Prediction/Testing

python main.py --in_seq_len IN_VALUE --out_seq_len OUT_VALUE --load_data --sequence_data --sequence_to_np

Training

python main.py --train_network --in_seq_len IN_VALUE --out_seq_len OUT_VALUE --load_data --sequence_data --sequence_to_np

Output

Output is stored as a MATLAB file in .mat file. Data is stored in the following dictionaries.

• date: Start date of the first day of the sequence. That is, the first date of the start of the input sequence.
• y_pred_mean: Mean prediction/inference output from different prediction samples: --n_samples VALUE. Dimension is (batch, out_seq_len, x_dim, y_dim)
• y_std_mean: Standard Deviation of the prediction/inference output from different prediction samples: --n_samples VALUE. Dimension is (batch, out_seq_len, x_dim, y_dim)
• y_target: Ground truth/Target values for the predicted output. Dimension is (batch, out_seq_len, x_dim, y_dim)