Find random values that match a specific criteria

I have a table with 2000 rows with the 8 columns collected from traffic data. However, the only columns I'm interested in are speed and time.

I have saved the time and speed columns as vectors. The time column format was in datetime but I have converted it to datenum to make it easier to work with.

time, speed
'2022-03-01 05:10:03', 55
.....

I want to find values of speed that are greater than the median speed value and then randomly keep only 20% of these values.

So far, I have attempted this:

speed_idx = find(speed > median(speed));% find the index of speed values > median speed

speed_idx_red = round((20/100)*length(speed_idx)); %keeping 20% of values > median speed

final_speed = speed(randperm(length(speed), speed_idx_red)); %randomised 20% of speed values

1. How can I simplify this? I think I'm finding the wrong final_speed, as I should perform the randomization first and then keep 20% of those values, but I'm not sure how to do that.
2. How can I find the time values that match the random speed values I've found?

Any help would be much appreciated!

Answer 1

[Fifteen12]

Broken down:

speed = randi(80, 10, 1);
m = median(speed);
pool = speed(speed > median(speed));
len = length(pool);
final_speed = pool(randi(len, floor(len * 0.2), 1));

Find the time values:

time = randi(100, 10, 1);
final_time = time(ismember(speed, final_speed));

Note that this method can give you more values for final_time than for final_speed if there are duplicates in final_speed. You'll have to choose how to handle duplicates. If you don't care about duplicates, and just want the random values, you can use unique to strip random numbers away from speed before finding the indices of time.

[newb_matlab help]

Can the speed be simplified to:

speed_median = speed(speed > median(speed));
final_speed = speed_median(randi(length(speed_median),floor(length(speed_median) * 0.2),1));

Also, for the time, you were right. I have a lot of duplicates and so the time vector is longer than the speed vector. I'll think of some ways to find the exact time values that match the random speed values. Thanks!

[Fifteen12]

Yes this works, as long as when you're searching for the time indices you use speed and not speed_median, otherwise your indices will be off.

You can use a unique function to strip the time variables, for instance

speed = randi(100, 10, 1);
time = randi(100, 10, 1);
time = time(ismember(speed, unique(speed)));
speed = unique(speed);

This removes all the duplicates speeds from time as well as speed. I think you'll need to do this before you do your median check (and randomization), otherwise you might call values that no longer exist.