Troubleshooting No Matches in Company Data Using Example Notebook and Exact Name Matching

Sandy C. · 2024-04-03T19:05:05.196Z

I have been playing with the example notebook and have tried using our company data as well as using the test.csv from the repo, and am not able to get any matches. Even when I specify the Name has to be EXACT for our company data. I have confirmed that I see some records in the company dataset where the name is an exact match. I have run through about 40 iterations and am not really getting any closer to a solution. Any thoughts/ideas on how I can move forward? Thanks, Sandy

Zingg Community

Troubleshooting No Matches in Company Data Using Example Notebook and Exact Name Matching | Zingg Community

11 comments

· Sorted by Oldest

Sonal G.
·
That’s strange. What is the size of data and labelDataSampleSize?
Sandy C.
·
The data set is 70995 I have the following configurations setup: args.setNumPartitions(20) args.setLabelDataSampleSize(0.01)
Sonal G.
·
Maybe try changing labelDataSampleSize to 0.1 and see if it gets better
Sandy C.
·
Thanks Sonal. I will try that again, and report back. I did try many different values and didn't seem to get a better response, but will definitely try again and let you know.
Sonal G.
·
Sure, please keep me posted Sandy. Also, when you say 40 iterations, is it number of records or number of times you ran findTrainingData? Mostly you will start seeing some matches post the 3rd or 4th round, so hang in there if thats the case.
Sandy C.
·
I have run at least 10 iterations with the same setup and have only seen 1 match. I must have something setup wrong because I am asking to have an exact match on name, and it is still suggesting things that don't match. Here is an example of what I see:
Sandy C.
·
Also, after each run I am printing out the n_pos and n_neg and they are always 0 even though I go through the match widgets and mark them as no match.
Sonal G.
·
Oh. It seems these are not getting saved then and hence no learning has happened. There is an explicit cell to save the labels. What happens when you label and run that?
Sandy C.
·
I think I found my problem. I have been going to the cell right after we setup the performance parameters and have selected "run all below". I just manually ran each cell, independently and the n_pos and n_neg are set correctly. I think by running the cell after the labeling and getting an error somehow interrupted the process once I actually labeled the data and ran the save again.
Sandy C.
·
I am so sorry to be bothering you with these user errors!! I really appreciate your time!!
Sonal G.
·
Happy to see this sorted Sandy C. and thanks for your continued effort on working with Zingg. 👍 😊 if you feel the instructions can be improved, please do consider sending a pull request

Sonal G.
·
That’s strange. What is the size of data and labelDataSampleSize?
Sandy C.
·
The data set is 70995 I have the following configurations setup: args.setNumPartitions(20) args.setLabelDataSampleSize(0.01)
Sonal G.
·
Maybe try changing labelDataSampleSize to 0.1 and see if it gets better
Sandy C.
·
Thanks Sonal. I will try that again, and report back. I did try many different values and didn't seem to get a better response, but will definitely try again and let you know.
Sonal G.
·
Sure, please keep me posted Sandy. Also, when you say 40 iterations, is it number of records or number of times you ran findTrainingData? Mostly you will start seeing some matches post the 3rd or 4th round, so hang in there if thats the case.
Sandy C.
·
I have run at least 10 iterations with the same setup and have only seen 1 match. I must have something setup wrong because I am asking to have an exact match on name, and it is still suggesting things that don't match. Here is an example of what I see:
Sandy C.
·
Also, after each run I am printing out the n_pos and n_neg and they are always 0 even though I go through the match widgets and mark them as no match.
Sonal G.
·
Oh. It seems these are not getting saved then and hence no learning has happened. There is an explicit cell to save the labels. What happens when you label and run that?
Sandy C.
·
I think I found my problem. I have been going to the cell right after we setup the performance parameters and have selected "run all below". I just manually ran each cell, independently and the n_pos and n_neg are set correctly. I think by running the cell after the labeling and getting an error somehow interrupted the process once I actually labeled the data and ran the save again.
Sandy C.
·
I am so sorry to be bothering you with these user errors!! I really appreciate your time!!
Sonal G.
·
Happy to see this sorted Sandy C. and thanks for your continued effort on working with Zingg. 👍 😊 if you feel the instructions can be improved, please do consider sending a pull request