Compute Accuracy for Pixel Classification Error 002517

[Ed_]

I am getting this error while the tool on a single classification 

ERROR 002517: This thematic raster has pixel values which are not in the attribute table.

However the pixel values are the same as the attribute table 

This error does not occur when running the tool on a multiclass prediction raster for example when the Class Name and Class Value are NWL and WL and 1 and 2 respectively. 

How can this this be fixed? Is the tool not designed for single class predictions?

[DanPatterson]

What is your nodata value?

Your question/post and this

BUG-000139826 for ArcGIS Pro

are the only two links that come up with reference to the error and the tool being used.

[Ed_]

Hi Dan, hope all is well and thank you for response, so the no data value was empty, so I changed it to 255 as mentioned in the bug. However, the error still persists.

Interestingly though, in this case as mentioned in the post (This error does not occur when running the tool on a multiclass prediction raster for example when the Class Name and Class Value are NWL and WL and 1 and 2 respectively.), the raster has empty no data value as well, but the tool runs fine with no such error. 

This tells me, there might be some other bug in the tool. 

[DanPatterson]

Tech Support is your best bet.

Although I can see how a multiclass prediction woud work for an input with a single class since there is nothing to assess the results of the prediction (eg mis-classifications etc)

[PavanYadav]

hi @Ed_ 

This error typically occurs because the accuracy assessment process requires a comparison between at least two classes to build a confusion matrix. If the raster is 'constant' (only contains one value) or contains background pixels not defined in the attribute table, the tool won't have the data it needs to complete the math. Adding a background class to your data might resolve the error and allow the tool to run as intended