How does DIADEM handle+INF, -INF and NAN's

[jgdub]

How does DIADEM handle+INF, -INF and NAN's

[Brad_Turpin]

Hi jgdub,

Assuming that you're talking about the usual IEEE denormals (and there are many of them) for NaN, +INF, -INF, then DIAdem maps all of these values to one IEEE denormal value in DIAdem which it displays and processes as DIAdem NoValues.  If you then save the data from DIAdem back to a data file, you get only NaN denormals, so you lose the +INF or -INF distinctions, if there were any.

Inside of DIAdem NoValues are handled comprehensively, flexibly and sensibly in graphing and analysis.  Were there any particular steps inside DIAdem you were curious about?

Brad Turpin
DIAdem Product Support Engineer
National Instruments

[jgdub]

Hi Brad,

I recently saw a demo of DIAdem and was quite impressed.

I work with data sets were missing data is represented by NaN (ie data were the instrument was offline), data over the upper detection limit is represented by +INF and data below the lower detection limit is represented by -INF. From what I saw the NaNs are handled as you say "comprehensively, flexibly and sensibly" but from your description the INFs are essentially ignored. I was hoping for something different in the way you handle INFs. I'm certain that I can find a workaround but it would be simpler for the INFs to be retained in the data sets and handled in calculations. Any chance for this to be implemented?

jgdub

[Brad_Turpin]

Hi jgdub,

I understand the appeal of retaining the +INF and -INF tags to those special denormals, since they have particular and useful meaning to you.  DIAdem is not going to do that any time soon, though.  What exactly would you like to see happen to these values?  If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

Brad Turpin
DIAdem Product Support Engineer
National Instruments

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First I

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First I would

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First I would like

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First I would like to

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.

[jgdub]

Hi Brad

Sorry for the delay in the conversation.

What exactly would you like to see happen to these values? 

First I would like to include

If we did have INFs in DIAdem, what would you do with them?  How would a +INF propogate in a smoothing function or a statistics function any differently than a NaN?  What would you like to see on graphs of your data when an INF occurs?

What do your data files look like, and how are you importing them into DIAdem at present?  It would be possible with a custom DataPlugin to map (during file import) the INF values to an appropriately +/- large real number or alternatively to create an additional "Rail" channel with [-1, 0, 1] values corresponding to whether each real value is [railed low, not railed, railed high].  Then I could imagine supporting all the graphing ideas you might have for your INF values.