Smile :)

Smile Classifier - The challenge is to create a classifier that can distinguish a smiling face.

Setup

Be sure to clone this repository and unzip the data on your AWS instance

cd ~
git clone https://github.com/lukas/smile.git
cd smile
unzip master.zip

Create an account at https://app.wandb.ai/login?invited if you don't have one. Copy an api key from your profile and paste it after calling wandb login below.

In your AWS terminal run:

pip install -r requirements.txt --upgrade
wandb login

Now try running the default scaffolding:

wandb run smile.py

Help

1. Slack Channel
2. Keras Docs
3. Previous Work
4. Other datasets
5. Data Augmentation
6. Transfer Learning

Files

• smile.py - scaffolding to get you started
• smiledataset.py - loads the data
• master.zip - the face dataset, smiles are in the positive directory the rest are in the negative directory

Stuck?

1. Ask for or help
2. https://blog.slavv.com/37-reasons-why-your-neural-network-is-not-working-4020854bd607

Extra Credit

Deploy the model.

Ideas

Simple, Effective:

• Normalize train_X and test_X
• Add dense layers
• Add convolutional layer(s)?
• Add dropout
• Change learning rate
• Experiment with activation functions

Fancier, Maybe Effective:

• Data Augmentation (https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html)
• Transfer Learning (https://www.learnopencv.com/keras-tutorial-transfer-learning-using-pre-trained-models/) Other ideas?

GPU on AWS

export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.1-cp27-none-linux_x86_64.whl
sudo pip install --upgrade $TF_BINARY_URL