Standard Tableau Workbook display bugs

[samaybar]

There are a number of minor display bugs in the Tableau workbook, although it is hard to identify whether they are a function of data or intrinsic to the workbook.

Some relate to filters maybe. 1) Share of voice only shows top 3 categories by default 2) Data Types show the category names twice 3) Age Gender tornados show "Gender Reversed" label at the bottom 4) Topics in Topic Exploration only show for the top 3 categories by default 5) The colors for each category are different from pane to pane

@haganbt I am happy to fix all of these things (and/or as well as the volume issue) if you'd like, except maybe the fifth one, which I am not sure I can figure out easily. Let me know if that would help!

[haganbt]

@samaybar - thanks for the feedback. Unfortunately there are intricacies within Tableau that complicate what we are trying to do and may force us to rethink the strategy. When Tableau is opened with a new data set, it will actually try and write data back to the workbook based on the new data set and whatever clicks are preformed e.g. what is selected as default. If you save the workbook thereafter, these changes are then saved and then can effect the next data set the workbook is used with.

It could be that you have opened the workbook that has previously had some additional state saved to it. I am also aware that the workbook within source control has some state pollution in there and when i find time to continue with this, I need to backtrack to find when that was introduced.

As of now, im not sure if Tableau can support what we are trying to do. My guess is that it can, but the workbook creation process needs to be 100% generic, and possibly copying it out with the source files on each pull is a workable solution.

I welcome any support in working on this process but until I look in to it further, you are likely to find such issues im afraid.

[haganbt]

@samaybar - I have done some further testing and found the issue. Whenever a filter is created, it writes the filter values to the workbook xml, which of course the values change as we swap out the data set. What this means is that if we do continue with this method, we can only create workbooks without filters i.e. the user cannot drill in to the data. This seems to significantly less value to me but something we can discuss within the SE team.

CC @tdbudden

[samaybar]

I'm not sure I totally follow. Can we work around this issue by placing the CSV files in a relative subdirectory rather than root and using a new copy of the template Tableau workbook each time?

Agree that the drill down seems like one of the great values of this whole thing! On Sep 2, 2015 10:09 PM, "haganbt" notifications@github.com wrote:

@samaybar https://github.com/samaybar - I have done some further testing and found the issue. Whenever a filter is created, it writes the filter values to the workbook xml, which of course the values change as we swap out the data set. What this means is that if we do continue with this method, we can only create workbooks without filters i.e. the user cannot drill in to the data. This seems to significantly less value to me but something we can discuss within the SE team.

CC @tdbudden https://github.com/tdbudden

— Reply to this email directly or view it on GitHub https://github.com/datasift/SE-PYLON-exporter/issues/52#issuecomment-137301564 .

[haganbt]

@samaybar - no, unfortunately not. Whenever we link two worksheets, we need to create a join at the data level and also create a filter i.e. whenever somebody clicks on an element, its filters the entire dashboard for that element. If we cannot filter the data, you cannot drill down e.g. click on the brand, then topic, then hashtag. All we can do is present separate visualizations side by side.

This is similar to the iPython approach where there is no interactivity, but if thats the case we may as well use a different tool that is more suited to static output.

[samaybar]

So maybe still being naive here -- it looks like it was filtering to 3 items by default -- is that because Automotive has 3 tags and that is what you created with? If you made the template using an index with 10 tags would it then work for anything with 10 or fewer? On Sep 2, 2015 10:20 PM, "haganbt" notifications@github.com wrote:

@samaybar https://github.com/samaybar - no, unfortunately not. Whenever we link two worksheets, we need to create a join at the data level and also create a filter i.e. whenever somebody clicks on an element, its filters the entire dashboard for that element. If we cannot filter the data, you cannot drill down e.g. click on the brand, then topic, then hashtag. All we can do is present separate visualizations side by side.

This is similar to the iPython approach where there is no interactivity, but if thats the case we may as well use a different tool that is more suited to static output.

— Reply to this email directly or view it on GitHub https://github.com/datasift/SE-PYLON-exporter/issues/52#issuecomment-137302479 .

[haganbt]

@samaybar - its a good thought but if we were to create an example with 10 items, it would hard code the names of those 10 items in to the xml, and hence have unpredictable results with a different data set as we are currently seeing.

[tdbudden]

This is unfortunate! It may be worth asking someone from Tableau if there's a workaround, or posting a question on the Tableau forum.

[scosden]

This may be oversimplifying things but is there a pathway here where we could dynamically write to the Tableau xml based on the dataset (CSV output) itself, such that each instance of the workbook is ostensibly re-built based on pre-defined dimensions (e.g. tags, topics, age/gender pyramids)?

On Wed, Sep 2, 2015 at 10:25 PM, haganbt notifications@github.com wrote:

@samaybar https://github.com/samaybar - its a good thought but if we were to create an example with 10 items, it would hard code the names of those 10 items in to the xml, and hence have unpredictable results with a different data set as we are currently seeing.

— Reply to this email directly or view it on GitHub https://github.com/datasift/SE-PYLON-exporter/issues/52#issuecomment-137303056 .

Scott Cosden Sales Engineer | DataSift

e: scott.cosden@datasift.com p: (347) 404-1995 t: @scottcosden https://twitter.com/scottcosden Learn more about DataSift https://datasift.com/

This email contains confidential information and is for the exclusive use of the addressee/s. If you are not the addressee, then any distribution, copying or use of this email is prohibited. If received in error, please advise the sender and delete it immediately.

DataSift, Inc | Office: DataSift, 157 Columbus Avenue, Suite 503, New York, NY 10023

[haganbt]

Hi @scosden - quite possibly. Looking at the Tableau docs however this is not documented and hence would need to be reverse engineered. Im looking in to it further before I make a final call but it might simply be not worth the effort.

[haganbt]

@samaybar @scosden - OK I think I may have found a fix. It is possible to add a filter that uses the "Use All" filter option which appears not to hard code any values back to the workbook. I have added a new test file "/tableau/standard-tableau-new.twb" which has two simple joins and filters for testing.

Please give it a test across a few different data sources and see if you can find any issues.

When you try and exit the workbook, it will prompt you to save - so make sure you say no, or pull from source again if you do accidentally overwrite the file.

It may end up being better for us to automatically copy the workbook and to output that on each export along with the source files.

FYI @tdbudden @sheanineseven

[scosden]

Thanks, Ben!

Great stuff. I will test this as soon as I have some time later in the weekend.

On Fri, Sep 4, 2015 at 7:53 PM, haganbt notifications@github.com wrote:

@samaybar https://github.com/samaybar @scosden https://github.com/scosden - OK I think I may have found a fix. It is possible to add a filter that uses the "Use All" filter option which appears not to hard code any values back to the workbook. I have added a new test file "/tableau/standard-tableau-new.twb" which has two simple joins and filters for testing.

Please give it a test across a few different data sources and see if you can find any issues.

When you try and exit the workbook, it will prompt you to save - so make sure you say no, or pull from source again if you do accidentally overwrite the file.

It may end up being better for us to automatically copy the workbook and to output that on each export along with the source files.

FYI @tdbudden https://github.com/tdbudden @sheanineseven https://github.com/sheanineseven

— Reply to this email directly or view it on GitHub https://github.com/datasift/SE-PYLON-exporter/issues/52#issuecomment-137879999 .

Scott Cosden Sales Engineer | DataSift

e: scott.cosden@datasift.com p: (347) 404-1995 t: @scottcosden https://twitter.com/scottcosden Learn more about DataSift https://datasift.com/

This email contains confidential information and is for the exclusive use of the addressee/s. If you are not the addressee, then any distribution, copying or use of this email is prohibited. If received in error, please advise the sender and delete it immediately.

DataSift, Inc | Office: DataSift, 157 Columbus Avenue, Suite 503, New York, NY 10023

[haganbt]

@samaybar @scosden - Gents, just FYI I have simplified the process in that a new Tableau workbook is copied to the output directory along with the source files. This should also help with supporting multiple data sets as the same source Tableau workbook will not be used across multiple source files.

I have updated the docs accordingly: https://github.com/datasift/SE-PYLON-exporter#tableau-workbook

Just need the platform to come back to life to test with other data sets!

FYI @tdbudden @sheanineseven

[haganbt]

@samaybar closing thread. All logged issues should be resolved except 5) as the color allocation across tabs cannot be guaranteed.