mgalarnyk / datasciencecoursera Goto Github PK

I guess there is something wrong in Stanford_Machine_Learning/Week6/MachineLearningSystemDesign.md. It shows that the statement "A good classifier should have both a high precision and high recall on the cross validation set." is False, however I think it should be True because we can see this statement from the course video. Moreover, I have tried it in the quiz, the result is as follow:

Matlab/Octave:

J = (1/(2*m)) *sum( (((X*theta)-y).^2))

Python :

s = np.power(( X.dot(theta) - np.transpose([y]) ), 2)
J = (1.0/(2*m)) * s.sum( axis = 0 )

They look equivalent except the python has that np.transpose([y])
Why is it needed?

BTW, My Octave version of this cost function is the same as yours.

This is probably not a bug, but it is confusing. You've done a nice job of doing the Python version. It would be an improvement to at least comment on that. I really wanted to do the assignment in NumPy, but Ng's tutorial on Matlab was so easy to follow that I just did the Octave version. Now I can compare the syntax. A Tabla Rosa!

pollutantmean <- function(directory, pollutant, id = 1:332) {

###Format number with fixed width and then append .csv to number
fileNames <- paste0(directory, '/', formatC(id, width=3, flag="0"), ".csv" )

###Reading in all files and making a large data.table
lst <- lapply(fileNames, data.table::fread)
dt <- rbindlist(lst)

if (c(pollutant) %in% names(dt)){
return(dt[, lapply(.SD, mean, na.rm = TRUE), .SDcols = pollutant][[1]])
}
}

###Example usage
pollutantmean(directory = '~/Desktop/specdata', pollutant = 'sulfate', id = 20)

**Q1: Please can you explain what have you done in highlighted portion (.SD and then .SDcols)?

Q2: Also this, .(n = .N) ??**
{--
complete <- function(directory, id = 1:332) {

###Format number with fixed width and then append .csv to number
fileNames <- paste0(directory, '/', formatC(id, width=3, flag="0"), ".csv" )

###Reading in all files and making a large data.table
lst <- lapply(fileNames, data.table::fread)
dt <- rbindlist(lst)

return(dt[complete.cases(dt), .(nobs = .N), by = ID])

}

###Example usage
complete(directory = '~/Desktop/specdata', id = 20:30)

--}

it is wrong. The learning rate &=0.3 still looks high compared with 0.1. The right answer is or should be; Rather than use the current value of α, it'd be more promising to try a smaller value of α (say α=0.1).

You have provided solutions for the course, I am really thankful for that. As I was doing the exercises , I found something worth mentioning to you. Week 2 , Question 4 ; the answer is different. Two options are correct. Please check for that. The transpose of v (1 cross 7) multiplied by w (7 cross 1) gives one number . Maybe they changed the options over time because this option was not present.
Again thanks for your work. I will see your content on YouTube too.

It would be very nice to get an explanation of the answer.

Seriously i am not able to classify the problem.

Why some time it goes to classification and some time regression.

Hi,

i have recently started working on prediction. please help me on how to prediction weather by using previous data (not from ) skymate/accueweather site.

pls post any queries
[email protected]

thanks
Venkat

Thank you for the great job with the course materials. I would like to reuse some of the quiz questions for my statistics course with attribution.

Please let me know if that is okay.

Maybe eventually you could add a CC-BY license if you want to encourage reuse?
https://github.com/santisoler/cc-licenses?tab=readme-ov-file#cc-attribution-40-international

Unfortunately, the information in the questionnaires is outdated. The questions have already been changed.