predictOneVsAll.m - Programming Assignment 3 Machine Learning

Key Takeaways

so we want to be able to predict for a one versus all logistic regression and so we've trained all ten different logistic regression classifier and we've trained them all for all of the ten different classes that we have and all of those weights are values trained parameters theta or stored in all theta so what we want is to generate prediction predictions virtual first of all which we use sigmoid for but we need to check first how should we multiply or theta and are our input X so we run this we see that we have X which is our input so we have 5000 examples of images and all of them have about 400 pixels and our parameter theta has 401 trained parameters and 10 for M for 10 different classifiers so with all of the 10-digit so we can see here is that we want to calculate these predictions to be a sigmoid of X times of theta transpose so if we run this and look at it we see that okay there's a lot of of them as if you look at the top for example our first image remember here that the first here the index 0 is a 0 and this is the image 1 image to etc up to 9 so for this one the probability or our prediction is quite high that this is a a 9 and similarly for many of these and then when we scroll down a little bit we get there ones etc so what we want now is that we want the maximum values from our predictions and that's the the final prediction that we make right now we have actually made ten predictions and all of them had different probabilities and we want to see well which one is the greatest so what we can use is matlab's max function so we take the maximum of the predictions and we do it I'm not sure what this does to be honest but it's in the documentation to use it but along the dimension for us to do we want to take it for each for one row look at all the columns to take maximum of those and that's the dimension to and we from that we get a max value and the indices for it and so what we can do is that we can check so Maxim box will be five thousand by one so we can take max well let's just you look at ten of them run it we get something like this and we recognize this one we can run it again just to make sure predictions get all of the 5000 and let's say we've taken out the ten first one so remember a ten here means that the number is a nine and here's the probability that we believe it's a nine so this one and we do that for all 5000 so after that what we do is that we just return the illness's so that's essentially what we care which which digit do we believe that this image is and see so yeah when we do that we get the training set accuracy of about a 95% which is yeah about what we should get so thanks for watching this video and see you in the next one