How do I calculate day difference using more than one date?

I have the following table:

Table A:

I want an output that looks like this:

1. Calculate the day difference between the transaction_date(where cancel_flag = No) and transaction_date(where cancel_flag = Yes).

2. If there's more than 1 cancellation_flag = Yes. The day difference used should be the minimum.

Thanks in advance,

For each record, the only 'cancel' rows you are interested in are the one just before or the one just after the current row when the data set is sorted by transaction_date. Because of this, solutions involving window functions seem quite appropriate here.

For any given row, you can get the date of the prior cancel transaction by

max(Case When Cancel_Flag='Yes' Then transaction_date End)
  Over (Partition By ID Order By Transaction_Date Rows Between Unbounded Preceding And Current Row)

, and the date of the following cancel transaction with

min(Case When Cancel_Flag='Yes' Then transaction_date End)
  Over (Partition By ID Order By Transaction_Date Rows Between Current Row And Unbounded Following)

Just use each in a datediff() with the current rows transaction date, and you've got two possible results that you can select from to get the final result.

Select ID,Transaction_Date,Cancel_Flag,
    Case When prior_cancel is null or next_cancel<abs(prior_cancel)
         Then next_cancel Else prior_cancel End as Days_Since_Cancel
From (
    Select A.*,
      datediff(day,Transaction_Date,
        max(Case When Cancel_Flag='Yes' Then transaction_date End)
         Over (Partition By ID Order By Transaction_Date Rows Between Unbounded Preceding And Current Row)
      ) as prior_cancel,
      datediff(day,Transaction_Date,
        min(Case When Cancel_Flag='Yes' Then transaction_date End)
         Over (Partition By ID Order By Transaction_Date Rows Between Current Row And Unbounded Following)
      ) as next_cancel
    From Table_A A
)
Order By ID,Transaction_Date

EDIT ADDITION
Note that, in place of min(...) you can use first_value(... Ignore Nulls) and in place of max(...) you can use last_value(... Ignore Nulls). These might be a tiny bit more efficient because while you cannot determine min & max without examining the entire window frame, in theory first and last can be determined without examining every element. These are always functionally equivalent when the Order By column and the min/max(column) are the same, in this case Transaction_Date.