random-sampling | Java 8 for the problem of random | Machine Learning library

I think the approach you've done creates a number 1:length(mtcars$gear) for each gear group so you will have repeat row numbers for each group. Then, when you subset it isn't working, see in your output above you have row number 7 in both gear group 3 and 4.

Base R

I would use split first to split by gear:

I think your rand() comparison is off:

I forgo efficiency for readability. Note that I generate a random choice for each row but only pick the number I need to fill in the nulls. Theoretically, I can make it such that I only pick random numbers for those missing values.

Your elasticsearch query is correct and you don't need scripts to perform what you want. It is just a problem with probabilities. For a short answer, replace the multiplier (i.e., field_value_factor) for the weight of 50 by 40 and the multiplier for the weight of 25 by 30 and you will get the expected result.

Basically, the problem is that multiplying a random value by a weight is not producing a weighted distribution where the weight is the multiplier. The multiplier can be derived from the weight, but there are not the same.

I can give you an example with your case. For the weight 50, if the random value is above 0.5, it will necessarily have the highest score (0.5 * 50 >= 1 * 25). Since a value of 0.5 as a probability of 50%, you now for sure that the item with weight 50 will be returned at least half of the time.

But even if the random value for weight 50 is below 0.5, it can still be selected. In fact its probability to be selected in this case is 1/3.

I'm just a bit surprised by your result because its probability should be more like 66% (i.e., 50% + 50%/3) and the other probabilities should be around 16.5%. Maybe try to increase the number of runs to be sure.

You do not need to compute the multiplier with this solution but you must provide a range, e.g., min_value and max_value for each document. max_value is the sum of min_value and the document wight and min_value is the cumulative sum of the weight of the previous documents.

If you have for example 4 documents with weights 5, 15, 30, 50, then the ranges could be :

• Documents with weight 5 : min_value = 0, max_value = 5
• Documents with weight 15 : min_value = 5, max_value = 5+15 = 20
• Documents with weight 30 : min_value = 20, max_value = 20+30 = 50
• Documents with weight 30 : min_value = 50, max_value = 50+50 = 100

The corresponding elasticsearch query is

Well, if you want to be able to write code like this:

Same logic like before when you do the sample , but here I convert between the binary and decimal twice, with unnesting , then join back the result

You can subset the data first by case ID == 1. To ensure occurrence of 1s and 0s, we use rep function and set replace to False in sample function.
Here's a solution.

I compared the two queries execution times using BigQuery standard SQL with the natality sample dataset (137,826,763 rows) and getting a sample for source_year column of size n. The queries are executed without using cached results.

Query1:

Instead of your runif line, use