Hi John, thanks for your comments. Responses below, hope they are helpful.
1 - it takes a lot of memory running about 200 datas with large records, takes very long to process, slows down computer dramatically. I only test on about 200 records.
RESPONSE: Yes, agreed, which is why you’d only use this tool as a last resort.
2 - the good things is while I was testing on 200 records I noticed a accuracy about 95%, but this may be on less complex records. Do you know if there is anything out there that will produce more accuracy? I say this because if it's not close to 100% there are chance where there could be error and risky to use especially with large records, unless we also do a manual check which you have to kind of check the whole records which is almost like doing it manually.
RESPONSE: controlled by the ‘Algorithm’ and ‘NFPercent’ parameters. You may also want to use FuzzyPercent to return the actual %age match. (but that’s even more processing)
Do you know if the match is more accurate with data that has less characters like 20 vs more characters like 100? Because I was doing a initial test and notice that it produce more accuracy with data with less characters or it may be because I was looking into a bigger file.
RESPONSE: The code will always match shortest string against the longest.
3 - if the data is looking for is not there, the match is completely off --- give random data from the look onto record which is inaccurate info.
RESPONSE:Controlled by ‘NFPercent’ parameter.
4 - it usually match on the first record it see, even though on the list there may be a closer match down further.
Response: The code will return the first highest match entry in the list. If subsequent highest entries are required, use the RANK parameter.
Do you know if there are any thing that is more accurate than this out there, and could this be converted into script? Also the ones that are not there should not give a random record that is completely off instead it should say n/a.
RESPONSE: Controlled by ‘NFPercent’ parameter.
over all it is good, but I still think it is risky depending on this b/c sometimes it could give you a match and it may not be correct, off course the chances are less but I think a check on the work after is needed but then is may seems a little manual.
RESPONSE: As with anything which matches two non-identical entries there is a risk that the entry returned is not the one you wanted it to. You CAN tailor it using the parameters ‘Algorithm’, ‘NFPercent’ and Rank, and that risk can be demonstrably quantified using the FuzzyPercent UDF against the lookup string and the returned entry.
></o
