Mandatory assignment 1

IN5400/IN9400 - Machine Learning for Image Analysis University of Oslo Spring 2019

Content

Everything you need for this exercise is contained in this folder. A brief description of the content follows.

Submission

Once you are done working, run the collectSubmission.sh script ( on Linux); this will produce a file called IN5400_assignment1.zip. On Windows, use e.g. 7-zip. Do not include the data directories in your zip file.

Then upload the zip-file file to devilry (devilry.ifi.uio.no). You can make multiple submissions before the deadline.

Good luck!

uio_in5400_s2019_mandatory1_assignment.ipynb

Everything related to the assignment. This should be self-contained, and all information is found in this notebook. You can start the notebook from the command line with

$ jupyter notebook uio_in5400_s2019_mandatory1_assignment.ipynb

Content of supplied code

The exercise contains this notebook and two folders with code: dnn and cnn:

 → tree
.
├── cnn
│   ├── conv_layers.py
│   └── __init__.py
├── dnn
│   ├── import_data.py
│   ├── __init__.py
│   ├── main.py
│   ├── model.py
│   ├── run.py
│   └── tests.py
├── figures
│   ├── backprop_conv.png
│   ├── cifar10_progress_default.png
│   ├── convolution_same.png
│   ├── convolution_same_x11.png
│   ├── convolution_same_x12.png
│   ├── convolution_same_x33.png
│   ├── mnist_progress_default.png
│   └── svhn_progress_default.png
├── README.md
└── uio_in5400_s2019_mandatory1_assignment.ipynb

dnn

This folder contains the whole dense neural network program. All functions that you are to implement in this exercise is found in dnn/model.py, but you are of course free to edit everything you want.

When you have implemented everything, you should be able to test your classifier with

$ python dnn/main.py

main.py

Handles program flow, data input and configurations. You should be able to run this file as an executable: $ python dnn/main.py.

You should not need to change anything here.

import_data.py

Handles import of the following three datasets

• MNIST
• CIFAR10
• SVHN

You should not need to change anything here.

run.py

Contains training and evaluation routines.

You should not need to change anything here

model.py

Implements all the important logic of the classifier.

Everything you need to implement will be located in this file.

tests.py

In this file, predefined arrays are defined. To be used when checking your implementations.

cnn

In this folder, code related to the convolutional neural network implementations reside.