observations.caltech101_silhouettes

caltech101_silhouettes(
    path,
    resolution=28
)

Load the Caltech 101 Silhouettes data set (Marlin, Swersky, Chen, & Freitas, 2010). It contains binarized 28x28 (or 14x14) pixel images of objects belonging to 101 categories. There are 4,100 examples in the training set, 2,264 examples in the validation set and 2,307 examples in the test set.

Args:

  • path: str. Path to directory which either stores file or otherwise file will be downloaded and extracted there. Filenames are caltech101_silhouettes_28_split1.mat if resolution is 28; else caltech101_silhouettes_14_split1.mat.
  • resolution: int, optional. Resolution of pixel images, 14 or 28.

Returns:

Tuple of np.ndarray tuples, (x_train, y_train), (x_test, y_test), and (x_valid, y_valid).

Marlin, B., Swersky, K., Chen, B., & Freitas, N. (2010). Inductive principles for restricted boltzmann machine learning. In Artificial intelligence and statistics.