Cropformer is a genomic selection method based on the architecture of Convolutional Neural Networks (CNNs) combined with a multi-head self-attention mechanism, which is used for crop phenotype prediction.

Python 3.10 / 3.11.

Optional: Hardware accelerator supporting PyTorch.

We provided a pre-packaged Conda environment for directly running Cropformer.

conda env create -f cropformer.yml

We provide a step-by-step guide for running Cropformer.

Convert the Ped file into a model-compatible CSV format. Test files are available in the data folder.

python ped_to_matrix.py

Upon successful execution, "Success" will be printed, and four files will be generated: X_train.csv, y_train, X_test.csv, and y_test.csv.

Tips： Please note the modification of the file path

Train your own Cropformer.

You need to provide four input files: X_train.csv, y_train.csv, X_test.csv, and y_test.csv.

python model.py

The model file is obtained after training is complete.

Tips： Please note the modification of the file path. Test files are available in the data folder.

python model_prediction.py

Tips： Please note the modification of the file path.

Please note that appropriate modifications should be made according to the model parameters of your own training.

Test files are available in the data folder.

Tips：Please note the modification of the file path. Test files are available in the data folder.

Any general-purpose computer that supports PyTorch can install this software, including systems such as Windows 10+, and Linux. On a Linux computer with 32GB of memory and a 16-core processor, the installation time for this software is less than 10 minutes. The running time is closely related to the dataset; generally, on computers with GPU support, training can be completed within 1 hour using datasets of a not too large scale.