Merge branch 'stl10'

Author: christianversloot

README.md

@@ -22,6 +22,7 @@ Hi there, and welcome to the `extra-keras-datasets` module! This extension to th
   * [KMNIST-K49](#kmnist-k49)
   * [SVHN-Normal](#svhn-normal)
   * [SVHN-Extra](#svhn-extra)
+  * [STL-10](#stl-10)
 - [Contributors and other references](#contributors-and-other-references)
 - [License](#license)
 
@@ -154,6 +155,18 @@ from extra-keras-datasets import svhn
 
 ---
 
+### STL-10
+The STL-10 dataset is an image recognition dataset for developing unsupervised feature learning, deep learning, self-taught learning algorithms. It contains 5.000 training images and 8.000 testing images, and represents 10 classes in total (airplane, bird, car, cat, deer, dog, horse, monkey, ship, truck).
+
+```
+from extra-keras-datasets import stl10
+(input_train, target_train), (input_test, target_test) = stl10.load_data()
+```
+
+<a href="./assets/stl10.png"><img src="./assets/stl10.png" width="100%" style="border: 3px solid #f6f8fa;" /></a>
+
+---
+
 ## Contributors and other references
 * **EMNIST dataset:**
   * Cohen, G., Afshar, S., Tapson, J., & van Schaik, A. (2017). EMNIST: an extension of MNIST to handwritten letters. Retrieved from http://arxiv.org/abs/1702.05373
@@ -162,6 +175,8 @@ from extra-keras-datasets import svhn
   * Clanuwat, T., Bober-Irizar, M., Kitamoto, A., Lamb, A., Yamamoto, K., & Ha, D. (2018). Deep learning for classical Japanese literature. arXiv preprint arXiv:1812.01718. Retrieved from https://arxiv.org/abs/1812.01718
 * **SVHN dataset:**
   * Netzer, Y., Wang, T., Coates, A., Bissacco, A., Wu, B., & Ng, A. Y. (2011). Reading digits in natural images with unsupervised feature learning. Retrieved from http://ufldl.stanford.edu/housenumbers/nips2011_housenumbers.pdf / http://ufldl.stanford.edu/housenumbers/
+* **STL-10 dataset:**
+  * Coates, A., Ng, A., & Lee, H. (2011, June). An analysis of single-layer networks in unsupervised feature learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics (pp. 215-223). Retrieved from http://cs.stanford.edu/~acoates/papers/coatesleeng_aistats_2011.pdf
 
 ## License
 The licenseable parts of this repository are licensed under a [MIT License](./LICENSE), so you're free to use this repo in your machine learning projects / blogs / exercises, and so on. Happy engineering! 🚀

assets/stl10.png

@@ -2,4 +2,5 @@
 
 from . import emnist
 from . import kmnist
-from . import svhn
+from . import svhn
+from . import stl10

extra_keras_datasets/__init__.py

@@ -0,0 +1,51 @@
+'''
+  Import the STL-10 dataset
+  Source: https://cs.stanford.edu/~acoates/stl10/
+  Description: The STL-10 dataset is an image recognition dataset for developing unsupervised feature learning, deep learning, self-taught learning algorithms.
+
+  ~~~ Important note ~~~
+  Please cite the following paper when using or referencing the dataset:
+  Coates, A., Ng, A., & Lee, H. (2011, June). An analysis of single-layer networks in unsupervised feature learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics (pp. 215-223). Retrieved from http://cs.stanford.edu/~acoates/papers/coatesleeng_aistats_2011.pdf
+
+'''
+
+from keras.utils.data_utils import get_file
+from scipy import io as sio
+import shutil
+import numpy as np
+
+def load_data(path='stl10_matlab.tar.gz'):
+  """Loads the STL-10 dataset.
+  # Arguments
+      path: path where to cache the dataset locally
+          (relative to ~/.keras/datasets).
+  # Returns
+      Tuple of Numpy arrays: `(input_train, target_train), (input_test, target_test)`.
+  """
+  path = get_file(path,
+                    origin='http://ai.stanford.edu/~acoates/stl10/stl10_matlab.tar.gz')
+
+  # Temporarily extract .tar.gz in local path
+  local_targz_path = './stl-10'
+  shutil.unpack_archive(path, local_targz_path)
+
+  # Load data from Matlab file
+  # Source: https://stackoverflow.com/a/53547262
+  train = sio.loadmat(f'{local_targz_path}/stl10_matlab/train.mat', )
+  test = sio.loadmat(f'{local_targz_path}/stl10_matlab/test.mat')
+
+  # Remove temporary file
+  shutil.rmtree(local_targz_path, ignore_errors=True)
+
+  # Define training data
+  input_train = train['X'].reshape((-1, 3, 96, 96))
+  input_train = np.transpose(input_train, (0, 3, 2, 1))
+  target_train = train['y'].flatten()
+
+  # Define testing data
+  input_test = test['X'].reshape((-1, 3, 96, 96))
+  input_test = np.transpose(input_test, (0, 3, 2, 1))
+  target_test = test['y'].flatten()
+
+  # Return data
+  return (input_train, target_train), (input_test, target_test)