Module 5: What are some examples of how intelligent computers process and display data?
Module 5 – High School
In this module, you will make use of the classical machine learning and deep learning techniques that you learned in the previous module to create predictive systems that make use of data pertaining to a Mars rover.
This notebook will be divided into two assignments that will require you to make use of your knowledge of the machine learning pipeline. The first assignment involves a dataset for logistic regression where you will make use of a dataset contained in a CSV file that contains information gathered about the Martian weather, such as temperature and atmospheric pressure. The second assignment involves using a dataset of satellite images of various Martian terrain features, which will therefore require the use of TensorFlow given that the most logical approach is a deep learning paradigm.
The objective of this module is to test your ability to effectively interpret a dataset such that you can train a model that is appropriate for the given task and test its ability to generalize. These procedures are indispensable for machine learning tasks; you will need to consider these approaches as you begin to devise ideas for a versalite project for the contest.
Part 1 – Logistic Regression and Temperature Controls
Part 2 – Image Classification for Martian Terrain
Part 1 – Logistic Regression and Temperature Controls
Suppose that you are helping to manage the Mars Exploration Rover. Given that the Martian atmosphere is significantly different than that of the Earth’s atmosphere, special measures must be taken to ensure that the rover can function effectively under extreme temperatures. In particular, the rover’s electronics and computer systems must remain intact under extreme conditions, lest the rover is unable to survive a sudden drop in surface temperature. Due to the red planet’s distance from the Sun, it is not uncommon to see differences in day and night temperatures of up to 113 degrees Celsius.
For more information on Martian weather conditions and the rover’s systems, please use the following link as a reference: https://mars.nasa.gov/mer/mission/rover/temperature/.
For more on Mars weather see this NASA website or this one
Module 5 Part 1- Assignment 1-Image Classification for Martian Terrain
Your task is to load a CSV file containing various features that pertain to atmospheric data gathered by the rover and use the dataset to train a logistic regression model. You will need to manually add a new a new column that will serve as the labels for the dataset, where “1” (the positive class) indicates that the rover must activate its temperature control systems and “0” indicates that the rover does not need to deploy these systems at that moment.
For the purposes of this task, minimum temperatures that are lower than -73 degrees Celsius will require the rover’s temperature control systems to be activated.
Consider the relevance of the features in the dataset to the task at hand. Not all of the features will be particularly useful for predicting whether temperature control should be used.
Remember to carry out the proper preprocessing steps, including addressing any missing values. Use this Jupyter Notebook to do the four assignments shown below.
Part 2 – Image Classification for Martian Terrain
The image above is an example found within a given training example in the dataset. Note that this training example is a result of data augmentation. Data augmentation is the artificial generation of new data from existing data, to train new machine learning (ML) models. Machine Learning models require incredibly large and varied datasets for initial training and finding these kinds of data sets in the real wold is difficult even for space data. Data augmentation artificially increases the dataset by making small changes to the original data either by a person or a computer. You can augment a dataset using a machine learning tool, for use in machine learning.
From the perspective of a rover, what could the satellite image above tell us about the Martian terrain? How can we make use of such data to facilitate navigation and avoid particular terrain types?
This assignment moves away from classical machine learning and will test your intuition for building and training a deep learning model that is capable of carrying out image classification with multiple classes.
The dataset with which you will be working contains a series of Martian terrain landmarks gathered from satellite data that are divided into several classes, which include the following:
- 0 – Other.
- 1 – Crater.
- 2 – Dark dune.
- 3 – Slope streak.
- 4 – Bright dune.
- 5 – Impact ejecta.
- 6 – Swiss cheese.
- 7 – Spider.
Your task will be to construct a model based on convolutional neural networks with TensorFlow such that it is capable of taking in a satellite image and determining the type of terrain displayed within the given input. In the context of this task, think about how this system could be used to aid a rover’s navigation through potentially hazardous terrain. Perhaps the predictions made by your model could be used as an input to a separate Python script that would relay a message that indicates the “level” of hazards that are present in a particular direction? What could a machine learning model do that current systems in place cannot do to make the best use of this information for navigational purposes?
Module 5 Part 2- Assignment 2 -Image Classification for Martian Terrain
As you complete this assignment, think about the potential advantages and disadvantages of using satellite data versus image data that is gathered directly from the rover’s perspective. The next module will expand upon this question and cover important information that pertains to the processes that are currently in place to allow for the exchange of such data between a rover and Earth.
Hint: Use the previous module as a reference, for you have already trained a model for multiclass classification with the MNIST dataset.
.
Make new document/Jupyter notebooks for each of these Assignments
Videos to support your Learning
Logistic regression
Neural Networks Part 1
Neural Networks Part 2
Neural Networks Part 3