What's the logic behind PERCENTILE.INC Excel function? - excel

I would like to know how does Excel think to calculate the values on the function PERCENTILE.INC. I'm making some studies on Percentile and Quartile, I got the below results:
How does Excel think to calculate the values on column F?
Here's the formulas I'm using:
=PERCENTILE.INC(B2:B21; 0,75) ==> F2
=PERCENTILE.INC(B2:B21; 0,50) ==> F3
=PERCENTILE.INC(B2:B21; 0,25) ==> F4
=PERCENTILE.INC(B2:B21; 0,00) ==> F5

Short answer - the position of a given percentile when the data is sorted in ascending order, using percentile.inc, is given by
(N-1)p+1
where p is the required percentile as a fraction from 0 to 1 and N is the number of points.
If this expression gives a whole number, you take the value at this position (e.g. percentile zero gives 1, so its value is exactly 22). If it's not a whole number, you interpolate between the value at the position given by the whole number part (e.g. for p=0.25 it's 5 and the value at this position is 52) and the value at the position one higher (in this case position 6 so the number is 55), then multiply the difference of the two values (3) by the fraction part (0.75) giving you 2.25 and finally add this to the lower of the two values giving you 54.25. A shorter way of saying this is that you go a quarter of the way between the two nearest values. So you have:
If you wished to show the logic as an Excel formula, you could implement the expression shown here on the right (where h, in the second column of the table, is the position calculated from the formula above and x is the value at that position)
like this:
=LET(p,J3,
range,I$2:I$21,
N,COUNT(range),
position,(N-1)*p+1,
lower,FLOOR(position,1),
fraction,MOD(position,1),
upper,CEILING(position,1),
lowerValue,INDEX(range,lower),
upperValue,INDEX(range,upper),
difference,upperValue-lowerValue,
lowerValue+fraction*difference)

Related

EXCEL - Dual VLOOKUP and Interpolation

I have a table on Excel with data as the following:
Meaning, I have different JPH based on the %SMALL unit and the number of active stations.
I need to create a matrix like the following (with %SMALL on horizontal and STATIONS on vertical axes):
And the formula for each cell should:
Take the input of Stations (column "B")
Check, for that specific Stations number, the amount of data on the other table (like make a filter on STATIONS for the specific number)
Perform an VLOOKUP for checking the JPH based on the %SMALL value on row 2
Interpolate for the exact JPH value, if not found on table
For now, I was able to create the last part (the VLOOKUP and the interpolation), with the following:
=IFERROR(VLOOKUP(C2;'EARLY-STATIONS'!$F:$H;3;FALSE);AVERAGE(OFFSET(INDEX('EARLY-STATIONS'!$H:$H;MATCH(C2;'EARLY-STATIONS'!$F:$F;1));0;0;2;1)))
The problem I'm facing is than with this, the calculation is not checking the number of stations, so the Iteration is not accurate.
Unfortunately I cannot use VBA macros to solve this.
Any clue?
This is an attempt because more clarity is needed in terms of all possible scenarios to consider, based on different input data and how to understand the "extrapolation" process. This approach understands as extrapolation the average of two values (lower and greater), but the idea can be customized to any other way to calculate it. Per tags listed in the question I assume there is no Excel version constraint. This is O365 solution:
=LET(sm, A2:A10, st, B2:B10, jph, C2:C10, smx, F1:J1, sty, E2:E4, NULL, "",
GETLk, LAMBDA(x,y,mode, FILTER(jph, (st=y)
* (sm = INDEX(sm, XMATCH(x, sm, mode))), NULL)),
GET, LAMBDA(x,y, LET(f, FILTER(jph, (jph=GETLk(x,y, 1))
+ (jph=GETLk(x,y, -1)), NULL), IF(#f=NULL, NULL, AVERAGE(f)))),
HREDUCE, LAMBDA(yi, DROP(REDUCE("", smx, LAMBDA(ac,x,
HSTACK(ac, GET(x, yi)))),,1)),
DROP(REDUCE("", sty, LAMBDA(ac,y, VSTACK(ac, HREDUCE(y)))),1))
The above formula spills the entire result, I don't think for this case you can use a LOOKUP-like function.
Here is the output:
The highlighted cells where the average is calculated.
Explanation
The main idea is to use DROP/REDUCE/HSTACK/VSTACK pattern to generate the grid. Check my answer to the following question: how to transform a table in Excel from vertical to horizontal but with different length on how to apply it.
We use two user LAMBDA functions to abstract some calculations:
GETLk(x,y,mode), filters jph name based on %SMALL and Stations columns values, based on input values x (x-axis value from the grid), y (y-axis value form the grid) respectively. The third input argument mode, is for doing the approximate search in XMATCH (1-next largest, -1 next smallest). In case the value exist in the input table, XMATCH returns the same value in both cases.
GET(x,y) has the logic to find the value or if the value doesn't exist to calculate the average. It uses the previous LAMBDA function GETLk. We filter for jph values that match the input values (x,y), but we use an OR condition in the FILTER (+), to select both lower or greater values. If the value exist, returns just one value otherwise two values are returned by FILTER (f). Finally if f is not empty we return the average, otherwise the value we setup as NULL.
HREDUCE: Concatenate the result by columns for a given row of the grid. Check the referred question for more information about it.

Excel Dynamic Array formula to create a running product of a column

I need to create a running product from a column of numbers (I could use a row, but a column is easier to demonstrate here.) The input might be any arbitrary array. In fact, in the application where I would deploy this, it will not be a range, but rather another dynamic array within a LAMBDA formula. Here is an example of the Input column of numbers and the desired Output from the formula:
Inputs
Expected Dynamic Array Output
10
10
8
80
3
240
4
960
5
4800
The formula would spill the results.
There are lots of solutions for a running total, but I've found no solution for a running product. I have tried a few different approaches, including SUBTOTAL and AGGREGATE with no success. I have also built a number of approaches that get the result, but are hard-coded to a fixed number of rows. I need the formula to adapt to any arbitrarily sized number of rows. The following formula is the closest I have gotten so far.
This LET formula delivers the result, but, as you can see is fixed to 5 rows:
=LET( a, {10;8;3;4;5},
v, SEQUENCE( ROWS(a) ), h, TRANSPOSE( v ),
stagr, (v - h + 1) * (v >= h),
m, IFERROR(INDEX( a, IF(stagr>0,stagr,-1), ), 1),
almost, INDEX(m,v,h) * INDEX(m,v,h+1) * INDEX(m,v,h+2) * INDEX(m,v,h+3) * INDEX(m,v,h+4),
result, INDEX( almost, , 1 ),
result )
The arbitrary array of numbers input is placed in the variable a.
The next step is to create some indexes that will be used to address these numbers: v is a sequence of vertical rows for each number in a and h is a the same sequence, but transposed into columns. stagr is an index matrix that is created from v and h that will later be used to address each item in a to form it into a multiplication matrix. If you replace the last result with stagr, you can see the shape of stagr. It just shifts a column down by one row until they are shifted all the way down.
Now we create the mulitplication matrix m using stagr by simply using INDEX, like this: INDEX(a,stagr). But this is not exactly what is needed because it takes the first row value (10) and replicates it because an INDEX of 0 is treated the same as 1. To get what we want, I forced an error by using and internal IF statement like this: INDEX( a, IF(stagr>0,stagr,-1) ) to replace the 0 results with -1. i.e. it will produce this:
Now, replace the errors with 1's by using IFERROR, so this explains how m is created and why. The result is a matrix like this:
and by multiplying m row-wise, we get the output we want, but this is where I fail.
For illustration, I created a variable almost that shows how I am trying to do a row-wise multiplication.
almost, INDEX(m,v,h) * INDEX(m,v,h+1) * INDEX(m,v,h+2) * INDEX(m,v,h+3) * INDEX(m,v,h+4)
You can see that I crudely multiplied one column times the next and the next... and using h + offset to get there. This produces the almost matrix and result just delivers the first column of that matrix, which contains the answer.
While an answer might be a good replacement for almost that would be dynamically sized, that is not my real question. I want a running product and I suspect that there is a wholly different approach than simply replacing my almost.
Just to be clear, the result must be a dynamic array that spills with no helper cells or CSE drag-down.
oh... and no VBA. (#stackoverflow - please add a no-VBA tag)
The only way I can find is to use DPRODUCT with OFFSET, but that requires a title row. It does not matter what is in the title row(it can even be empty), just that it is included.
=DPRODUCT(OFFSET(A1,0,0,SEQUENCE(COUNT(A:A),,2)),1,$ZZ1:$ZZ2)
The $ZZ1:$ZZ2 can be any empty cell reference.
If the values in A are dynamic then we can do:
=DPRODUCT(OFFSET(A1,0,0,SEQUENCE(ROWS(A2#),,2)),1,$ZZ:$ZZ)
There are plenty of interesting answers here. But, if summation is easy why not take logarithms of the number you want to multiply, sum those logarithms and then calculate the exponent of your sum to return to the product of the original numbers.
i.e. exploit the fact that ln(a * b) = ln(a) + ln(b)
Whilst not available to everybody (yet) we can use SCAN()
Formula in A1:
=SCAN(1,{10,8,3,4,5},LAMBDA(a,b,a*b))
The 1st parameter is our starting value, meaning the 1st calculation in the nested LAMBDA() is '1*10'.
The 2nd parameter can both take a 1D- & 2D-array (written or range-reference).
The 3rd parameter is a nested LAMBDA() where the result of our recursive function will then be used for the 2nd calculation; '10*8'. And the 3rd...etc. etc.
In the above sample a vertical array is spilled but when horizontal input is used this will obviously result in an horizontal spilled output. When a 2D-array is used this will spill a 2D-array as result.

Excel: How to find closest number in table, many times

Excel
Need to find nearest float in a table, for each integer 0..99
https://www.excel-easy.com/examples/closest-match.html explains a great technique for finding the CLOSEST number from an array to a constant cell.
I need to perform this for many values (specifically, find nearest to a vertical list of integers 0..99 from within a list of floats).
Array formulas don't allow the compare-to value (integers) to change as we move down the list of integers, it treats it like a constant location.
I tried Tables, referring to the integers (works) but the formula from the above web site requires an Array operation (F2, control shift Enter), which are not permitted in Tables. Correction: You can enter the formula, control-enter the array function for one cell, copy the formulas, then insert table. Don't change the search cell reference!
Update:
I can still use array operations, but I manually have to copy the desired function into each 100 target cells. No biggie.
Fixed typo in formula. See end of question for details about "perfection".
Example code:
AI4=some integer
AJ4=MATCH(MIN(ABS(Table[float_column]-AI4)), ABS(Table[float_column]-AI4), 0)
repeat for subsequent integers in AI5...AI103
Example data:
0.1 <= matches 0
0.5
0.95 <= matches 1
1.51 <= matches 2
2.89
Consider the case where target=5, and 4.5, 5.5 exist in the list. One gives -0.5 and the other +0.5. Searching for ABS(-.5) will give the first one. Either one is decent, unless your data is non-monotonic.
This still needs a better solution.
Thanks in advance!
I had another problem, which pushed to a better solution.
Specifically, since the Y values for the X that I am interested in can be at varying distances in X, I will interpolate X between the X point before and after. Ie search for less than or equal, also greater than or equal, interpolate the desired X, then interpolate the Y values.
I could go a step further and interpolate N - 1 to N + 1, which will give cleaner results for noisy data.

Excel comparison with multiple IFs

I have a scale, based on which I decide the value of the coefficient for the multiplication. The scale looks as following:
Which means that:
for Category1: when value>=1.000.000 then coef is 1, when value>=500.000 then coef is 0.8 and etc.
Same logic applies for Category2;
Then I have input data in the following format:
Company !MainCat|Sales Amount|
Company1|T1 | 6.500.000|
Company2|T2 | 70.000|
I need to find corresponding coefficient, ratio of the coeffitient and the value (=ratio*MaxCoef). Currently, I am finding coef the following way:
- for company1:
=IF(C8>=$D2;$D$1;IF(C8>=$E2;$E$1;IF(C8>=$F2;$F$1;IF(C8>=$G2;$G$1;IF(C8>=$H2;$H$1;IF(C8>=$I2;$I$1))))))
That is literally hardcoded and doesn't look good. Maybe there is a better way of doing ? Any suggestions?
Formula view:
You can COUNTIF(range, [criteria] < value) * 0.2 as your add 0.2 per coef stage.
To you data do: =COUNTIF(D2:H2, "<"&C8) * 0.2, count how many stages the value passes * the value per stage.
Your count if range needs to be until H2 as I2 is 0, so inferior to value and gets counted.
To combine the COUNTIF() with a dynamic search for the right category based on MainCat you can MATCH() the MainCat with Code which will give the row where the Code is located and utilize INDIRECT() to apply it as range.
=COUNTIF(INDIRECT("D"&MATCH(B8,B:B,0)&":H"&MATCH(B8,B:B,0)),"<"&C8)*0.2
MATCH(B8,B:B,0) - will match the value on B8 (lets say T1) and return the row 2.
INDIRECT("D"&MATCH(B8,B:B,0)&":H"&MATCH(B8,B:B,0) = INDIRECT("D"&2&":H"&2) - will turn the text into an actual range to be use by the COUNTIF().
Create a table ‚Mapping’ that contains two columns, ‚Category’ and ‚Coefficient‘, then use INDEX-MATCH on it as described in https://www.deskbright.com/excel/using-index-match/.
=INDEX(Mapping[Category]; MATCH([Coefficient]; Mapping[Coefficient]; -1))
This example assumes that you put this formula into a table that has a column named ‚Coefficient‘ with the input value to your multiple IFs.
The trick is that as a match_type argument, provide either -1 or 1, according to your needs.
You can do this in VBA. Write your own function which ends in something like that
=MyOwnScale(C8; B8; A2:I3)
The first parameter of your VBA-function is the value, the second the category and the third is the range with the thresholds. So you can move your cascading IF-loops in VBA-Code and you (and your users) see only a clean function call in the cell.

Binning in Excel

Which formulae in MS Excel can we use for -
equi-depth binning
equi-width binning
Here's what I used. The data I was binning was in A2:A2001.
Equi-width:
I calculated the width in a separate cell (U2), using this formula:
=(MAX($A$2:$A$2001) - MIN($A$2:$A$2001) + 0.00000001)/10
10 is the number of bins. The + 0.00000000001 is there because without it, values equal to the maximum were getting put into their own bin.
Then, for the actual binning, I used this:
=ROUNDDOWN(($A2-MIN($A$2:$A$2001))/$U$2, 0)
This function is finding how many bin-widths above the minimum your value is, by dividing (value - minimum) by the bin width. We only care about how many full bin-widths fit into the value, not fractional ones, so we use ROUNDDOWN to chop off all the fractional bin-widths (that is, show 0 decimal places).
Equi-depth
This one is simpler.
=ROUNDDOWN(PERCENTRANK($A$2:$A$2001, $A2)*10, 0)
First, get the percentile rank of the current cell ($A2) out of all the cells being binned ($A$2:$A$2001). This will be a value between 0 and 1, so to convert it into bins, just multiply by the total number of bins you want (I used 10). Then, chop off the decimals the same way as before.
For either of these, if you want your bins to start at 1 rather than 0, just add a +1 to the end of the formula.
Best approach is to use the built-in method:
http://support.microsoft.com/kb/214269
I think the VBA version of the addin (step 3 with most versions) will also give you the code.
Put this formula in B1:
=MAX( ROUNDUP( PERCENTRANK($A$1:$A$8, A1) *4, 0),1)
Fill down the formula all across B column and you are done. The formula divides the range into 4 equal buckets and it returns the bucket number which the cell A1 falls into. The first bucket contains the lowest 25% of values.
General pattern is:
=MAX( ROUNDUP ( PERCENTRANK ([Range], [TestCell]) * [NumberOfBuckets], 0), 1)
You may have to build the matrix to graph.
For the bin bracket you could use =PERCENTILE() for equi-depth and a proportion of the difference =Max(Data) - Min(Data) for equi-width.
You could obtain the frequency with =COUNTIF(). The bin's Mean could be obtained using =SUMPRODUCT((Data>LOWER_BRACKET)*(Data<UPPER_BRACKET)*Data)/frequency
More complex statistics could be reached hacking around with SUMPRODUCT and/or Array formulas (which I do not recommend since are very hard to comprehend for a non-programmer)

Resources