In this assignment you will get familiar with standard multi-layer perceptrons. You will start with a multinomial logistic regression classifier and then convert it to an one-layer neural network. Then you will add more layers to the network including different types of activation function.
For those who are not familiar with python it is highly recommended to check https://docs.scipy.org/doc/numpy-dev/user/quickstart.html.
Also, check an ipython tutorial http://nbviewer.jupyter.org/github/jupyter/notebook/blob/master/docs/source/examples/Notebook/Notebook%20Basics.ipynb.
You should use python 2.7 for this assignment. Please, don't use python 3.x beacause there are some issues with pickle.
Files of this practical were updated for those students who have problems with downloading the data. Explicitly run the script cifar10/get_cifar10.sh. If it doesn't work, please, download the data and extract it manually.
The assignment consists of two parts: an ipython notebook python_assignment_1.ipynb and a pdf file paper_assignment_1.pdf with exercises that you will have to solve with pen and paper. You should follow the instructions in these files. You should work on both parts of the assignment.
For easier submission procedure we provide a simple script collect_assignment.py that will collect your submission files into a zip file that you should send to us. You can run the command as follows:
python collect_assignment.py --last_name your-last-name
To make sure you and we don't miss any important files or answers, please use the collect_assignment.py script. In case, however, that the script does not work for you, you can manually collect the files and the results. In that case the zip archive that you will send us must have the following structure:
lastname_assignment_1.zip │ python_assignment_1.ipynb │ python_assignment_1.pdf │ paper_assignment_1_solved.pdf │ └───uva_code │ │ __init__.py │ │ layers.py | | losses.py | | models.py | | optimizers.py | | solver.py
Please make sure that the structure of the file is correct, as the assignments are mostly going to be automatically checked and the specified structure is required. After you create the zip files, please send it to email@example.com ONLY. Because of the large number of students if you send it to any other email, we cannot guarantee that it will be corrected in time.
The deadline for the assignment is the 9th of November, 23:59.