Hi Pravin. I am glad to help. At some stage I play to move from feature selection to automatic feature engineering. In my experience it boosts the performance many times. If you are subscribed, you will get notified.
@@enlightenment609 Thanks I was able to understand the entire process and successful run the code. The only thing is my code tend to generate the same iteration result in each run.
@@tusuubiraedward9600 There can be many reasons. If the problem is too easy, population size is too small, or the random seed is the same then you will get the same answer. For the encoded problem, it is likely to be too easy: Picking from 4/5 attributes is not hard for a large enough population. Try some other problem where number of features is larger.
Thank you very much.. Please excuse me. I would like you if you would explain to me the possibility of modifying the chromosome.. I am working on a master's thesis in the field of predicting missing values. Can you help me with a code that predicts them? Or you can give me a reference that I can benefit from.. and I repeat my thanks to you
Your comment does not clarify why and how do you want to modify the chromosome just to predict missing values (from what?). ) If you want to build a classifier that predicts if a dataset/record has missing values then GA must only select features for such a classifier. In this case the same binary chromosome applies.
if you have already selected 5 features using genetic algorithm, then you must already have written a fitness function. Did you use a classification algorithm with the GA?
Yes, so just like the video uses linear Regression inside the fitness function, you can I stead call logistic Regression inside it. In fact the drop box code gives you a file called class feature fitness.R that uses classification with GA. Only it uses decision tree instead of logistic Regression.
Hi Luis. The code provided in the description beneath the video also contains the code to access the dataset. It is all self contained. Let me know if you didn't follow that.
Yes, that's very easy. You set the chromosome length = 23 in runGA.R. Then set the fitness function up to use the right dataset as well. See examples in class-feature-fitness.R.
You mean feature engineering with Genetic Programming? Yes, I have it in the plans. And in my experience it works very well. I have done it with Grammatical Evolution and it is a lot easier to do than with Genetic Programming.
You can search for GA toolkits in Python. You must then write a fitness function similar to what you have seen here and attach it to the Python code. For example, you can search for DEAP toolkit.