Embedded Files
This competition is very accessible! It is very easy to start and has no hardware requirements! Anyone from first year onwards can build something that works!
This competition is very accessible! It is very easy to start and has no hardware requirements! Anyone from first year onwards can build something that works!
Prizes!
Prizes!
There are 3 cash prizes for the best scores achieved on the test set of the competition!
The competition will run from your first accessing this page to the 30th of July 30th of August at noon PST after this point no more submissions will be considered.
1st - 100$
1st - 100$
2nd - 50$
2nd - 50$
3rd - 25$
3rd - 25$
Summary!
Summary!
Anyone can compete on a level playing field. There are no hardware requirements – see instructions and rules!
The goal of this competition is to get students invested and involved in their first machine learning project!
We will be evaluating everyone's skills in generating a classification model for the Fashion MNIST dataset!
The Fashion MNIST dataset is 10 classes with thousands of examples for each class, but all the images are only 28 by 28 pixels!
This will all be done in a very memory-limited environment, with only 2Gigs of memory ever available for training and testing!
(See the rules for more)
Instructions and Rules!
Instructions and Rules!
If you have any questions contact Declan at contact@declanmcintosh.com!
You may work in teams of any size!
We will be evaluating all models that you make ourselves on the Fashion MNIST dataset, the solutions you submit must be written in python.
Your final solution must be able to run in this browser Jupyter notebook website from mybinder, memory is very limited, that's the challenge! You add in your code and test it out!
Your solution must download the dataset from the official Keras download, you do not necessarily have to use Keras for your model or training, you just have to load your data from there (see the example code at the bottom of the page for how to do this).
Your submission will be run by the organizers in the mybinder Jupyter environment to ensure that it can run on the limited resources.
You can submit any number of times, the score will be updated based on your submissions and added to the scoreboard on this page!
Your submissions must
Run functionally the online Jupyter environment
Load the Fashion MNIST dataset from the official Keras download
Train and validate the model from scratch on only the train data
Test your model on the test data
Report your model's final accuracy
Submitted in a .txt file
Be emailed to ieeesb@uvic.ca
Have your names and V#s in the email
Example and Starter Code!
Example and Starter Code!
If you are looking for some information on how to use the online notebook environment or some ML information in general check out our old skillDev video below!
First, open an online Jupyter environment
Now copy and paste in the bellow starter code and you can test out the results. The results will be VERY bad but it should run!
You should in a few seconds see an output something along the lines of:
Train: X=(60000, 28, 28, 1), y=(60000,)
Test: X=(10000, 28, 28, 1), y=(10000,)
690/15000 [>.............................] - ETA: 4:04 - loss: 0.5903 - accuracy: 0.0922
Now here is the starter code!
# install packages!pip install keras!pip install tensorflow# import and download datafrom keras.datasets import fashion_mnistimport numpy as np(trainX, trainY), (testX, testY) = fashion_mnist.load_data()
trainX = np.expand_dims(trainX.astype('float16') / 255., axis=3)testX = np.expand_dims(testX.astype('float16') / 255., axis=3)
print('Train: X=%s, y=%s' % (trainX.shape, trainY.shape))print('Test: X=%s, y=%s' % (testX.shape, testY.shape))
# import some stufffrom keras.models import Sequential from keras.layers import Dense, Input, Flattenfrom keras.losses import binary_crossentropyfrom tensorflow.keras.utils import to_categoricalfrom keras.optimizers import Adam
# reformat labelstrainY = to_categorical(trainY)testY = to_categorical(testY)
# deine number of classes classNum = 10
# make modelmodel = Sequential()model.add(Flatten())model.add(Dense(1024,activation='softmax'))model.add(Dense(512,activation='softmax'))model.add(Dense(classNum,activation='softmax'))
model.compile(loss=binary_crossentropy, optimizer=Adam(), metrics=['accuracy'])
# define some hyper parameters and trainmodel.fit(trainX, trainY, batch_size=4)score = model.evaluate(testX, testY, verbose=0)print("Test accuracy: ", score[1])
Current Leaderboard
Current Leaderboard
Randon Abrams - Accuracy 0.9035
Liam Shatzel - Accuracy: 0.9033
Ridley Roelfs - Accuracy: 0.8762
Sample Code - Accuracy: 0.1016
If you want some help check out the previous IEEE skillDev!
Google Sites
Report abuse