CODEWITHSUNDEEP
pythonmachine-learningneural-networkjupyter-notebookconv-neural-network
user5492930
user5492930

Reputation:

Keras: My model trains without any given labels. How's it possible?

Yesterday I created my first convolutional neural network in keras but I forgot to add labels and it trained ... somehow, and I have no idea how. Can someone explain how the heck it trained without any labels given?

Jupyter Notebook: http://nbviewer.jupyter.org/github/getrasa/Jupyter-Notebook-Share/blob/master/Untitled.ipynb

Folder structure files/ train/ dogs/ dog.0.jpg ... cats/ cat.0.jpg ... files/ validation/ dogs/ dog.1301.jpg ... cats/ cat.1301.jpg files/test/ (1-13images).jpg

Upvotes: 1

Views: 139

Answers (1)

Lukasz Tracewski
Lukasz Tracewski

Reputation: 11377

No magic here. From the docs:

flow_from_directory(directory): Takes the path to a directory, and generates batches of augmented/normalized data. Yields batches indefinitely, in an infinite loop.

Arguments:

directory: path to the target directory. It should contain one subdirectory per class. Any PNG, JPG or BMP images inside each of the subdirectories directory tree will be included in the generator

As long as your data is split into subdirectories that correspond to your classes, ImageDataGenerator will produce labels out of these. For instance, take this directory structure:

train/
   cat/
   dog/
   eel/

The flow_from_directory will by default take these as categorical and use one-hot encoding in the background. That's how you get labels.

One final note: since you have only two classes, you can consider changing class_mode to binary.

Upvotes: 1

Related Questions