# Homework 1

In this homework, we will train a simple deep network to classify images from SuperTuxKart.

This assignment should be solved individually. No collaboration, sharing of solutions, or exchange of models is allowed. Please, do not directly copy existing code from anywhere other than your previous solutions, or the previous master solution. We will check assignments for duplicates. See below for more details.

You might need a GPU to train your models. You can get a free one on google colab. We provide you with a ipython notebook that can get you started on colab for each homework.

If you’ve never used colab before, go through colab notebook (tutorial)
When you’re comfortable with the workflow, feel free to use colab notebook (shortened)

Follow the instructions below to use it.

• Select the upload tab then select the .ipynb file.
• Follow the instructions on the homework notebook to upload code and data.

### Starter code and dataset

The starter code for this assignment can be found here. The starter code contains several useful scripts:

• bundle.py will zip up your homework, ready for submission. Only submit zip files created by this bundling script
• grader locally grades your submission, works with both zip files and your homework directory.

The starter code also contains a data directory where you’ll copy (or symlink) the SuperTuxKart classification dataset. Unzip the data directly into the homework folder, replacing the existing data directory completely. Make sure you see the following directories and files inside your main directory

homework
bundle.py
data
data/train
data/valid


You will run all scripts from inside this main directory.

As a first step, we will need to implement a data loader for the SuperTuxKart dataset. Complete the __init__, __len__, and the __getitem__ of the SuperTuxDataset class in the utils.py.

• The __len__ function should return the size of the dataset.

• The __getitem__ function should return a tuple of image, label. The image should be a torch.Tensor of size (3,64,64) with range [0,1], and the label should be int.

• Labels and the corresponding image paths are saved in labels.csv, their headers are file and label. There are 6 classes of objects. Make sure label background corresponds to 0, kart is 1, pickup is 2, nitro is 3, bomb is 4 and projectile 5.

Once you finish, you can visualize some of the images by their classes using:

python3 -m homework.visualize_data data/valid


Hint: We recommend using the csv package to read csv files and the PIL library (Pillow fork) to read images in Python.

Hint: Use torchvision.transforms.ToTensor() to convert the PIL image to a pytorch tensor.

Hint: You have (at least) two options on how to load the dataset. You can load all images in the __init__ function, or you can lazily load them in __getitem__. If you load all images in __init__, make sure you convert the image to a tensor in the constructor, otherwise, you might get an OSError: [Errno 24] Too many open files.

python3 -m grader homework -v


## Linear Model (15 pts)

Implement the LinearClassifier class in models.py. Define the linear model and all layers in the __init__ function, then implement forward. Your forward function receives a (B,3,64,64) tensor as an input and should return a (B,6) torch.Tensor (one value per class). You can earn these full credits without training the model, just from the correct model definition.

python3 -m grader homework -v


## Classification Loss (10 pts)

Next, we’ll implement the ClassificationLoss in models.py. We will later use this loss to train our classifiers. You should implement the log-likelihood of a softmax classifier.

$$-\log\left(\frac{\exp(x_l) }{ \sum_j \exp(x_j)} \right),$$ where $x$ are the logits and $l$ is the label. You may use existing pytorch functions to implement this.

python3 -m grader homework -v


## Training the linear model (30 pts)

Train your linear model in train.py. You should implement the full training procedure

• create a model, loss, optimizer
• load the data: train and valid
• Run SGD for several epochs
• Save your final model, using save_model

You can train your network using

python3 -m homework.train -m linear


Hint: You might find it useful to store optimization parameters in the ArgumentParser, and quickly try a few from the command-line.

Hint: You might find it useful to allow training of an existing model to continue. Use the torch.load function for that. See load_model.

Hint: Try to write your training code model agnostic. We will swap out the model below.

python3 -m grader homework -v


## MLP Model (30 pts)

Implement the MLPClassifier class in models.py. The inputs and outputs to the multi-layer perceptron are the same as the linear classifier. However, now you’re learning a non-linear function.

You can train your network using

python3 -m homework.train -m mlp


Hint: This part might require some tuning of your training code. Try to move most modifications to command-line arguments in ArgumentParser

Hint: Use ReLU layers as non-linearities.

Hint: Two layers are sufficient.

Hint: Keep the first layer small to save parameters.

You can test your trained model using

python3 -m grader homework -v


### Relevant Operations

python3 -m grader homework -v


will run a subset of test cases we use during the actual testing. The point distributions will be the same, but we will use additional test cases. More importantly, we evaluate your model on the test set. The performance on the test grader may vary. Try not to overfit to the validation set too much.

## Submission

Once you finished the assignment, create a submission bundle using

python3 bundle.py homework [YOUR UT ID]


python3 -m grader [YOUR UT ID].zip


We will use an automated grader through canvas to grade all your submissions. There is a soft limit of 5 submisisons per assignment. Please contact the course staff before going over this limit, otherwise your submission might be counted as invalid.

The online grading system will use a slightly modified version of python and the grader:

• Please do not use the exit or sys.exit command, it will likely lead to a crash in the grader
• Please do not try to access, read, or write files outside the ones specified in the assignment. This again will lead to a crash. File writing is disabled.
• Network access is disabled. Please do not try to communicate with the outside world.
• Forking is not allowed!
• print or sys.stdout.write statements from your code are ignored and not returned.

Please do not try to break or hack the grader. Doing so will have negative consequences for your standing in this class and the program.

## Honor code

This assignment should be solved individually.

What interaction with classmates is allowed?

• Talking about high-level concepts and class material
• Talking about the general structure of the solution (e.g. You should use convolutions and ReLU layers)
• Looking at online solutions, and pytorch samples without directly copying or transcribing those solutions (rule of thumb, do not have your coding window and the other solution open at the same time). Always cite your sources in the code (put the full URL)!
• Using any of your submissions to prior homework
• Using the master solution to prior homework
• Using ipython notebooks from class

What interaction is not allowed?

• Exchange of code
• Exchange of architecture details
• Exchange of hyperparameters
• Directly (or slightly) modified code from online sources
• Any collaboration
• Putting your solution on a public repo (e.g. github). You will fail the assignment if someone copies your code.

Ways students failed in past years (do not do this):

• Student A has a GPU, student B does not. Student B sends his solution to Student A to train 3 days before the assignment is due. Student A promises not to copy it but fails to complete the homework in time. In a last-minute attempt, Student A submits a slightly modified version of Student B’s solution. Result: Both students fail the assignment.

• Student A struggles in class. Student B helps Student A and shows him/her his/her solution. Student A promises to not copy the solution but does it anyway. Result: Both students fail the assignment.

• Student A sits behind Student B in class. Student B works on his homework, instead of paying attention. Student A sees Student B’s solution and copies it. Result: Both students fail the assignment.

• Student A and B do not read the honor code and submit identical solutions for all homework. Result: Both students fail the class.

## Installation and setup

### Installing the dependencies

Install all dependencies using

python3 -m pip install -r requirements.txt


Note: On some systems, you might be required to use pip3 instead of pip for python 3.

If you’re using conda use

conda env create environment.yml


The test grader will not have any dependencies installed, other than native python3 libraries and libraries mentioned in requirements.txt. This includes packages like pandas. If you use additional dependencies ask on piazza first, or risk the test grader failing.

### Manual installation of pytorch

Go to https://pytorch.org/get-started/locally/ then select the stable Pytorch build, your OS, package (pip if you installed python 3 directly, conda if you installed Anaconda), python version, cuda version. Run the provided command. Note that cuda is not required, you can select cuda = None if you don’t have a GPU or don’t want to do GPU training locally. We will provide instruction for doing remote GPU training on Google Colab for free.

### Manual installation of the Python Imaging Library (PIL)

The easiest way to install the PIL is through pip or conda.

python3 -m pip install -U Pillow


There are a few important considerations when using PIL. First, make sure that your OS uses libjpeg-turbo and not the slower libjpeg (all modern Ubuntu versions do by default). Second, if you’re frustrated with slow image transformations in PIL use Pillow-SIMD instead:

CC="cc -mavx2" python3 -m pip install -U --force-reinstall Pillow-SIMD


The CC="cc -mavx2" is only needed if your CPU supports AVX2 instructions. pip will most likely complain a bit about missing dependencies. Install them, either through conda, or your favorite package manager (apt, brew, …).