/corn_yield_prediction_mlcas

solution for MLCAS 2024 Corn Yield Prediction hosted on EvalAI

Primary LanguagePython


  • Although this solution only stopped at top 10 in the LB, here is a quick note for feature selection that was used, which can be used as a reference for similar task:

    • Raw 6-band satellite data, resized to 24x24, normalized to [0,1]
    • Location encoding
    • Nitrogen level encoding
    • Passing days from the planting day
  • I tested several models, but finally used a simple CNN taking one time-point data of a location as input, while using average of yield scores on all the time points as output for one location. The final model is here

  • Several ideas that were tested but were not showing good results for my experiments:

    • cnn-lstm
    • contrastive learning to initialize weights for the models

🎛 Development environment


mamba env create --file environment.yml
mamba activate corn_yield

💎 References