Homework 3: KNN and Neural Networks Solution

Before you start Homework 3, make sure you have finished the preparation step to set up the

environment. Note that you may use different operating systems (Windows, Mac or Linux). The

setup instuctions are different. If you have problems in this step, please come to TAs for help.

Code of HW3 has been tested to be compatible with python version 3.6 (it may work with other

versions of 3.X, but we won’t be officially supporting them.). You will need to make sure that during

your virtual environment setup that the correct version of python is used. If you would like to use

Python 2.7, it is also OK but keep in mind that you may have problems or errors, which may take

extra efforts.

Set up Python 3 (For Mac OS/Linux)

Most of you have already installed Python 3 for previous assignments. If you have done that, you

can skip this step. It is recommended to install Anaconda to manage Python and packages. Other

than that, for Mac OS X, you can also install Python 3 via Homebrew and call brew install

python3 . For Linux, you can install through sudo apt-get .

Set up Virtual Environment (For Mac OS/Linux)

In case you have different python package version and other issues, we recommend to use Python

virtualenv which sets up a vitural environment. At a high-level, a virtual environment creates an

environment where you may install packages that aren’t installed on your main system Python.

This is useful for separating out projects. Further, if you accidentally break your Python during

some package installation (it can happen and is not fun), but you were in a virtual environment, you

can simply delete the virtual environment and start a new. Use pip to install it: sudo pip

install virtualenv .

Please run the following in the terminal to start visual environment and install required packages of

HW3 after installation:




d HW3

virtualenv -p python3 .env

# Create a virtual environment (python3)

# Note: you can also use "virtualenv .env" to use your default python

source .env/bin/activate # Activate the virtual environment

pip install -r requirements.txt # Install dependencies (this takes a while)

# Work on the homework for a while ...

deactivate

# Exit the virtual environment

Alternatively, you can use Anaconda to set up virtual environment if you have installed it. To set up

a virtual environment with Anaconda, run (in a terminal):

conda create -n .env python=3.6 anaconda

# Create an environment called

.env

source activate .env

# Activate and enter the

environment

# Work on the homework for a while ...

source deactivate .env

# Exit the virtual environment

Check here for more details on managing virtual environments with Anaconda.

Do I have to use virtual environment? The answer is "No, probably you don’t". Be advised,

however, that if you choose not to use virtual environment, it is up to you to make sure all

dependencies in the code are installed globally on your machine. However, if things break, it is not

possible for us to help debug each student’s unique installation, as each computer setup is

different, and the bug may be any package you have previously used or installed interacting

negatively. That is why we prefer to have you use virtual environments and have provided you the

requirements.txt file for the homework.

Installing Python and virtualenv for Windows

Download Python package from https://www.python.org/downloads/release/python-364/ and

install it. Run cmd as administrator and run these commands:

pip install virtualenv

cd HW3/

virtualenv -p path to python3/python.exe .env # This creates the virtual

environment

.env\Scripts\activate.bat # This activates the virtual environment

pip install -r requirements.txt # This installs all the packages

# Do some work...

.env\Scripts\deactivate.bat # This deactivates the virtual environment

Note: We find the setup instructions from online (but all TAs do not have Windows sorry). Please tell us if

you have problems. We suggest to search on StackOverflow or post your problem on Piazza first.Set up and Launch Jupyter Notebook

To finish this homework, we use jupyter notebook for coding and report. An IPython notebook

lets you write and execute Python code in your web browser. IPython notebooks make it very easy

and interactive to tinker with code and execute it in steps. Check here for more details if you are

not familiar with Jupyter Notebook and how it works.

If your virtual environment installed correctly, you should NOT have to manually install it from the

instructions on the website. Just remember to run source .env/bin/activate in your

homework folder.

Run:

cd HW3/

jupyter notebook

Also you can use jupyter lab which is basically the same thing as jupyter notebook but

probably a little bit more convenient .

This will launch a new browser window (or a new tab) showing the notebook dashboard. Or you

can visit http://localhost:8888/ to open the dashboard.

When started, the jupyter notebook can access only files within its start-up folder (including any

sub-folder). Make sure your relevant files are on the desired path.

Note: To write down the answers in the notebook, you may need a little knowledge of markdown

syntax, which is basically like txt but more flexible and readable. Here is a brief tutorial. (Fun fact:

You can even use it when posting on Piazza!)

Congratulations for completing the setup for your coming exciting homework! Now let's start! ^_^

Data Preparation

Download the CIFAR-10 dataset (file size: ~163M). Run the following from the HW3 directory:

cd cs145/datasets

./get_datasets.sh

Also you can download the dataset here, and put it under cs145/datasets folder.

After downloading the dataset, you can start your notebook from the HW3 directory.

Task 1: KNN (30 points)

Task: Complete the k-nearest neighbors Jupyter notebook including codes and questions (following

the instructions and notes in knn.ipynb ). The goal of this workbook is to give you experiences

with the CIFAR-10 dataset, training and evaluating a simple classifier, and k-fold cross validation.

Print out the entire workbook and submit jupyter notebook together with all source codes.Task 2: Two-layer Neural Networks (70 points + 20 bonus points)

Task: Compete the two-layer neural network Jupyter notebook including codes and questions

(following the instructions and notes in toy_nn.ipynb ). Print out the entire workbook and submit

jupyter notebook together with all source codes.

Submission

Step 0: Make sure to finish the missing lines and questions in notebook.

Step 1: Convert your ipynb files to PDF files. If you have problems to do that from the

Jupyter Notebook in the browser, try to convert all your ipynb notebook files to HTML by

running ipython nbconvert --to html FILE.ipynb and print as PDF.

Step 2: Confirm that you have required files for submission: two python files ( knn.py and

neural_net.py ) and two notebook files ( knn.ipynb and toy_nn.ipynb ), which have been

converted to PDF already (you may name them as knn.pdf and toy_nn.pdf ).

Step 3: Submit a zip file of your homework to CCLE using zip_hw3.sh . Make sure you have

include all your code ( .py files) and notebook. Note: DO NOT submit dataset files.

Step 4: Check again your files can be sucessfully unzipped and click "Submit for Grading" in

CCLE. If you do not click "Submit for grading" and we probably notice a late submission

in your submission, we will apply late-time penalty. No email submission will be

accepted this time.

Grading

Grading policy of HW3: Code implementation (approx. 40%) + Results and Analysis (approx.

60%) in the notebook (excluding bonus questions) Points will be significantly deducted if you

only provide results with no required analysis in the notebook!

Points will be significantly deducted if you only provide results with no required analysis in the notebook!

Keep the academic honor code in mind.

With its status as a world-class research institution, it is critical that the University uphold the highest standards

of integrity both inside and outside the classroom. As a student and member of the UCLA community, you are

expected to demonstrate integrity in all of your academic endeavors. Accordingly, when accusations of

academic dishonesty occur, The Office of the Dean of Students is charged with investigating and adjudicating

suspected violations. Academic dishonesty, includes, but is not limited to, cheating, fabrication, plagiarism,

multiple submissions or facilitating academic misconduct.

Notes (Keep Updating)

1. Be aware of your installation if you have Python 2 and Python 3 at the same time on your

machine with diffrent python and package file path.

2. Pay attention to the indentation of code. Different code editors may have different default

indentation settings. Please make sure this does not bring errors or affect your code before

you start.

3. If you installed homebrew previously and then updated your OS system, you may have touninstall and reinstall homebrew.

4. If you are working in a virtual environment on OSX, you may potentially encounter errors with

matplotlib. Check here.

5. We know it is not that easy to fill in other's code. You are free to change the code we provide

even though it is not inside the missing lines. Keep in mind that you should not change the

whole structure of the codes and we will not support the such change on code.

6. Do NOT use other functions (such as kNN or dense layer in sklearn, tensorflow or pytorch) in

your inplementation. If you report your results by using these function directly, you will get 0

points.

7. More reminders to be updated!

Acknowledgement

We thank Prof. Jonanthan Kao, Zheng Cheng, Tianwei Xing, together with Serena Yeung & Justin

Johnson, for permission to use their code written for the ECE239AS: Neural Networks & Deep

Learning (Winter 2018). This graduate course is highly recommended and helpful if you are

interested in advanced techniques in neural networks.