The following notebooks are tutorials, demonstrating how easy it is to do deep learning with nolearn/lasagne library. Some basic knowledge of deep learning is assumed. You should know multilayer perceptions and convolutional neural networks. See http://deeplearning.net/tutorial/ or cs231n for an introduction.
- MinimalLasagneMLP.ipynb nbviewer shows how to create a multilayer perceptron.
- MinimalLasagneCNN.ipynb nbviewer shows how to create a (simple) convolution neural network
- DataAugmentation.ipynb nbviewer shows how to do training data augmentation (in principle)
- DataAugmentationII.ipynb nbviewer shows how to use training data augmentation to ease overfitting and do a better prediction.
Slides of a tutorial on these notebooks given at the Zurich Machine Learning Meetup can be found here
To install the required python packages follow the installation procedure descriped in https://github.com/dnouri/nolearn
To clone the demo itself
git clone https://github.com/oduerr/dl_tutorial.git
To start the ipython notebook server
ipython notebook
Alternatively you can use a VM. For example http://datasciencetoolbox.org/ provides a VM and Amazon AMIs. To install the data science toolbox and lasagne on top of it see README_DataScience_ToolBox.md for a step-by-step instruction.
Creating this tutorial I was very much inspired by the following great tutorials:
-
danielnouri describing the use of Lasagne in facial keypoint detection. A great tutorial which inspired this tutorial.
-
msegala Code Plankton quite simple code, which also a contains a
Batchiterator
- Lasagne Github Page good starting point, see also the mailing list
- Winning Solution for the plankton challenge from the creator of lasagne