Skip to content

Unconventional use of CNNs to process structured data (bidimensional panel data: multimodal time-series) and predict next day's weather summary with 96% Accuracy. McGill MMA

Notifications You must be signed in to change notification settings

carlosfg97/WeatherCNN_WarutNet

Repository files navigation

WeatherCNN Warut-Net

  • Simple Convolutional Neural Network architecture to predict the weather summary of the next day (Foggy, Breezy, Clear, Overcast, Mostly Cloudy, Partial Cloudy) based on a current day weather information. Data is pivoted into a matrix form (or panel data points) so the CNN is able to convolute across two dimensions and find local regions of relevant correlated information, similar to how an image processing would happen.

  • Python Notebook

  • Short video walkthrough of the notebook is available here

drawing

  • Accuracy
Training Set Test Set
96% 96%
  • Fast training times ( <60 seconds to fit 800K data points) using Tensorflow and Keras backend in a GPU-enabled Colab environment

  • CSV file uploaded for reference

About

Unconventional use of CNNs to process structured data (bidimensional panel data: multimodal time-series) and predict next day's weather summary with 96% Accuracy. McGill MMA

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published