Your mission is to train a neural network to steer a car using only image from a single camera. sample_student.py contains the basic functions you'll need to modify. First you'll need to create a method to train your neural network:
def train(path_to_images, csv_file):
'''
First method you need to complete.
Args:
path_to_images = path to jpg image files
csv_file = path and filename to csv file containing frame numbers and steering angles.
Returns:
NN = Trained Neural Network object
'''
Your method should return a trained neural network class. An example class (from the Neural Networks module) is included in sample_student.py. Your submitted script should also inlcude a NeuralNetwork() class - you are free to modify this class or write your own from scratch. You'll also need to create a predict method:
def predict(NN, image_file):
'''
Second method you need to complete.
Given your trained neural network class, and an image filename, load image, make and return single predicted steering angle in degrees, as a float32.
'''
Your code will be evaluated using evaluate.py, or a very similar variant. You can use this script locally to ensure you don't burn through your autograder submissions. Performance will be evaluated by comparing your predicted steering angles to human steering angles:
Due to limited compute capacity, your code will only be allowed to run for 10 minutes. Keep this in mind during development! There's lots of training strategies that take way too long, this time limit will force you to find nice fast training strategies (a useful skill, I promise). Finally, you may want to add your own timer to training to ensure you get as many training steps as you can without exceeding the time limit. If you exceed the time limit, your code will be termintated, and your submission will be scored as if your code fialed to run.
Download training data here.
For this challenge you are only permitted to use numpy, opencv, tdqm, time, and scipy. Tensorflow and other high level ML tools are not permitted.
| RMSE (degrees) | Points (10 max) |
|---|---|
| RMSE <= 12.5 | 10 |
| 12.5 < RMSE <= 15 | 9 |
| 15 < RMSE <= 20 | 8 |
| 20 < RMSE <= 25 | 7 |
| 25 < RMSE <= 30 | 6 |
| 30 < RMSE <= 40 | 5 |
| RMSE > 40, or code fails to run | 4 |