Observations after my first KNIME workflow

I've just finished my first non-trivial KNIME workflow and I had a few observations/requests. It may sound like a long list of complaints, but I am very impressed with KNIME. I plan to integrate it into a number of processes in the near future. Nice tool!!

Let me know if I can clarify any of the details.

-Kirk

* As I've mentioned previously, it would be VERY useful to have a parameter node in which variables affecting the workflow could be set. At this point I have 21 nodes, 6 of which require customized settings, and one of those is code related. Once I distribute this to co-workers, it will require that they understand exactly where adjustments need to be made and could be tricky.

* Scaling of chemical structures in an Interactive Table could be improved. It seems I have to scale the structure up to ~8 cm X 8 cm in order to get an acceptable structure depiction. Anything less seems to have colliding characters or bonds.

* The chemical structure depiction in PDF Report is very poor and often clipped. (I have not tried HTML Report.)

* When I have multiple view open on the data, e.g. Interactive Table and Scatter Plot, having the two views linked with respected to highlighted points is very useful. However, I have to go through a lot of clicks to get the desired effect: Select Point to UnHiLite, Click HiLite on menu, Select UnHighLight Selected, Select Point to HiLite, Click HiLite menu, Select HiLite Selected. Six clicks per change.

* Once a point is HiLite'd on a ScatterPlot (for example), there is no way to go directly to that point in the Interactive Table. A Scroll to Next HiLite would be VERY useful. At this point I have to scan through the table page by page hoping not to miss it.

Hi Kirk,

Just some comments on the "chemical" issues, you mentioned:

KirkD wrote:
* Scaling of chemical structures in an Interactive Table could be improved. It seems I have to scale the structure up to ~8 cm X 8 cm in order to get an acceptable structure depiction. Anything less seems to have colliding characters or bonds.

Yes, that's a known issue. We are currently relying solely on CDK to do the rendering, and their renderer is not the best one. Some of the CDK guys have written an improved renderer for SWT (Eclipse's windowing toolkit), however, KNIME's table are Swing, so I have the feeling, that it might not be that easy to integrate. We will have a look at it after our workshop ;-)

KirkD wrote:
* The chemical structure depiction in PDF Report is very poor and often clipped. (I have not tried HTML Report.)

Yep, I think that is related to CDK renderer, too, because the images in the PDF and the HTML reports are created by exactly the same components as in the tables. There is however another thing, that parts of the boundaries are sometimes drawn, which makes the images even uglier. We will also have a look at it.

KirkD wrote:

* Once a point is HiLite'd on a ScatterPlot (for example), there is no way to go directly to that point in the Interactive Table. A Scroll to Next HiLite would be VERY useful. At this point I have to scan through the table page by page hoping not to miss it.

Maybe the option "Show hilited only" will help you a bit in this case?

Regards,

Thorsten

Quote:
* When I have multiple view open on the data, e.g. Interactive Table and Scatter Plot, having the two views linked with respected to highlighted points is very useful. However, I have to go through a lot of clicks to get the desired effect: Select Point to UnHiLite, Click HiLite on menu, Select UnHighLight Selected, Select Point to HiLite, Click HiLite menu, Select HiLite Selected. Six clicks per change.

Did you encounter the context menu which opens up on right mouse click? Although it doesn't reduce the number of clicks, at least the ways you have to move with the mouse are considerably shorter.

You find the "show hilited only" in the "Hilite" menu of the InteractiveTable.

Regards,

Fabian Dill

Thanks for the responses.

Thor - I thought CDK might be to blame for the rendering. Not an easy fix, for certain. I know in Windows you can do scaling of a pre-rendered image - I'm not familiar with Swing to know if something of that sort is possible. Of course that would likely pixelate the image on reduction.

I will check out the Show HiLited only - I hadn't noticed that option.

Dill - yes, the context menu shortens the mouse mileage, but I'm still left with a lot of clicks. A standard approach is for each subsequent selection to cancel the last unless the Ctrl key is held down. At least that was the behavior I expected without thinking too much about it.

As I mentioned before, I don't want to come across as a complainer. These were just a few details I noticed after building my workflow. Overall the results are great and I'm excited to see what other tasks I can capture in KNIME.

-Kirk

Quote:
A standard approach is for each subsequent selection to cancel the last unless the Ctrl key is held down. At least that was the behavior I expected without thinking too much about it.

We decided to differentiate between selection and hiliting. As you may have noticed, the selection is only local while the hiliting is global (propagates to all views). Since with large datasets this may take some time, we wanted the user to hilite data points explicitely and not with the same behaviour as the selection.
There is a difference between selection an hiliting and for us it seemed ok to reflect this difference also in the way it is triggered.

That's the reason why it behaves not as selection. Hope you don't mind ;o)

Regards,

Fabian

It merely represents a small departure from what I have come to accept as "standard." It's a bit of a long trip to change highlighting. Perhaps my click finger endurance will improve over time. 8^)