How to learn Machine Learning in 4 days.
This is an absurd task, I am aware of that. Why I know its possible is because myself along with some other exception students Hack Reactor spent 12-16 hours day for 4 days to accomplish this. Below is a basic plan we came up with in order to help anyone else start what we did.
I would not have been able to do this by myself, and I would recommend at least 4 people try this together. You will need to motivate each other, split up the learning and teach each other along the way.
One of the harder things to get this up and running is the environment with all the tools you need. I have a post on how to Set up a Machine Learning environment.
The H/R team: Me, Andrew Spade, Kyle Geppert, Felix Trippier, Max Malin, Zak Zibrat, Guy Morita, Tuhin Chakraborty, Bill Shelton
How does your spam filter predict what emails to mark as spam? How does Google guess the traffic to the Giant’s game next week? How does Netflix make you movie recommendations? Machine learning is the art of using data to make predictions.
In this sprint you will be using records of survivors from the Titanic disaster and machine learning to create an algorithmic model capable of predicting whether a passenger lived or died, and what kind of things about a person would make them more or less likely to have survived ( for example, as women and children were given priority seating among the ship’s scarce lifeboats, they were much more likely to survive than other demographics).
Getting familiar with ML - Titanic survival prediction
is: using algorithms to analyze and train a large set of data to make predictions
is not: writing sentient AI which teaches itself
Understanding the basic concepts
In this assignment, you’ll learn how to apply machine learning to predict who will lived and died on Titanic voyage. You will start using simple pivot tables and regression in excel. Then using Python to apply regressions and implementing one of the random forest algorithms.
Watch Introduction to ML video
Watch the videos in the Introduction section of the coursera ML course
Learn basic terminology (and make sure to look up any other words you don’t know along the way)
Training / Validation / Testing Data
R (data analysis, stats, graphics, machine learning prototyping)
Rattle (a GUI library for machine learning in R)
Understand the data by using excel, google spreadsheet or a similar tool (pivot tables can be used for this). Determine:
Percent of people who survived by sex
Percent of people who survived by class
Percent of people who survived by age
Any other information you think may be relevant or that will help you understand the data
Load rattle and get comfortable with its interface, choose a model
Load the data into rattle
partition the data into training / validation / test sets
choose the target column (data you want to predict)
Visualize the data using rattle’s tools on the explore tab
Explore the options for each section
at a minimum use the Explore → Interactive → Latticist
Try to find a view to support the conclusions you came up with during the problem analysis section above
Explore the models available in rattle
Why are there different types of models?
When would you use each type?
Run multiple models and look at the data
understand what AUC (area under the curve) means
compare different model’s AUCs to see which may be better for your data
Choose a model to evaluate your data with
Once you have chosen a model, evaluate how well the model works on your validation data
Become familiar with the Evaluate Tab
In the Data section, choose validation. This will run your model on the validation data
Evaluate the testing data using the model you chose above
Choose type: Score, Data type: csv: test.csv
this will output a file in CSV format.
compare this to the answer.csv file to check the accuracy of your model.
Watch the coursera ML Regularization (week 3) - The problem of overfitting (10 min)
you can ignore the mathematical equations in the video, but you should understand the high level concept of overfitting and how to fix it
Watch the coursera ML Advice for Applying Machine Learning (week 6) - Model Selection and Train / Validation / Test sets (12 min)
understand why you need train / validation / and test data sets and how they help solve the overfitting problem
Kaggle TItanic - all the tutorials
Research a layout of the Titanic and use it to engineer at least two completely unique features
Go through the Scikit-learn tutorial
Check out NLTK - Natural Language Toolkit
