Skip to content

Latest commit

 

History

History
48 lines (32 loc) · 5.09 KB

README.md

File metadata and controls

48 lines (32 loc) · 5.09 KB
Raw

MakeHealth datathon [Website]

1. Dengue satellite data

Tabular data for Dengue can be downloaded in the following items

  1. Download

2. Satellite images feature extraction with deep learning models

Satellite features for 164 images for Medellin can be downloaded here in pickle format. The shape for each file is given by the structure (164, 100), where 100 represents the number of features obtained for each model.

  1. Features 1 - Download: Feature extraction variation based on resnet50 pretrained on Imagenet - Extracted from Sentinel 2 in Medellin between 2015-2018
  2. Features 2 - Download: Feature extraction variation based on vision transformers pretrained on Imagenet - Extracted from Sentinel 2 in Medellin between 2015-2018

For all models, find labels here: Download

3. Satellite images dimensionality reduction with Variational Autoencoders

Satellite embeddings for 164 images of Medellin can be downloaded here in pickle and csv format.

  • pickle: The shape for each pickle file is given by the structure (164, n_features), where n_features represents the number of features obtained for each model.
  1. Features 4 - Download: Embeddings generated using a variational autoencoder with latent space of 100 (100 features) in pickle format - Extracted from Sentinel 2 in Medellin between 2015-2018
  2. Features 5 - Download: Embeddings generated using a variational autoencoder with latent space of 200 (200 features) in pickle format - Extracted from Sentinel 2 in Medellin between 2015-2018
  • csv: The shape for each csv file is given by the structure (164, n_features + 1), where n_features represents the number of features obtained for each model and the extra column is the date of the image.
  1. Features 6 - Download: Embeddings generated using a variational autoencoder with latent space of 100 (100 features) in csv format - Extracted from Sentinel 2 in Medellin between 2015-2018
  2. Features 7 - Download: Embeddings generated using a variational autoencoder with latent space of 200 (200 features) in csv format - Extracted from Sentinel 2 in Medellin between 2015-2018 For all models, find labels here: Download

4. Satellite images dimensionality reduction with principal component analysis (PCA)

Satellite embeddings for 164 images of Medellin can be downloaded here in pickle and csv format.

  • pickle: The shape for each pickle file is given by the structure (164, n_features), where n_features represents the number of features obtained for each model.
  1. Features 8 - Download: Embeddings generated using the first 100 principal components in pickle format - Extracted from Sentinel 2 in Medellin between 2015-2018
  2. Features 9 - Download: Embeddings generated using the first 10 principal components in each band (120 features in total per image) in pickle format - Extracted from Sentinel 2 in Medellin between 2015-2018
  • csv: The shape for each csv file is given by the structure (164, n_features + 1), where n_features represents the number of features obtained for each model and the extra column is the date of the image.
  1. Features 10 - Download: Embeddings generated using the first 100 principal components in csv format - Extracted from Sentinel 2 in Medellin between 2015-2018
  2. Features 11 - Download: Embeddings generated using the first 10 principal components in each band (120 features in total per image) in csv format - Extracted from Sentinel 2 in Medellin between 2015-2018

For all models, find labels here: Download

5. Dates

Array of tahes in pickle format of shape (164):

  1. Dates - Download: Dates of each satellite image in pickle format.

6. Labels for ResNet50 and ViT Model

  1. Dataframe - Download: Labels of each images based on number of cases in CSV format.