Requirements:

• Python 2.7.1
• TensorFlow 1.4.1
• CIFAR-10 dataset downloaded from https://www.cs.toronto.edu/~kriz/cifar.html

In order to train the model, navigate to the root directory and run the desired .sh file located in the scripts folder (i.e scripts/conv_net.sh). Both bash files contain the following customizable options:

• data_path (String): The path to the folder containing the CIFAR-10 data from the root directory
• output_dir (String): The path to wherever you want the outputs stored
• log_every (int): The number of episodes between calculating validation score
• num_epochs (int): The number of epochs to train
• train_batch_size (int): The size of the training batch at each step
• eval_batch_size (int): The size of the evaluation batch when calculating validation accuracy
• l2_reg (float): The rate of L2 regularization
• learning_rate (float): The learning rate

Allowing the convolutional model to fully train without changing the parameters should result in a final accuracy of aroun 80% on the CIFAR-10 dataset.