How to create Empty Nominal values?

[Stormdamage]

Hello everyone, I am trying to use the GeneralizedSequentialPatterns (GSP) Weka node for Knime (https://nodepit.com/node/org.knime.ext.weka37.associator.WekaAssociatorNodeFactory%23GeneralizedSequentialPatterns%20(3.7))

When reading the paper for this Node, the input should be lists of variable lengths. In Knime these lists need to be split into columns for processing by the GSP node, which means some of them have missing values, which the Node won’t accept. How can I turn a missing value into an empty nominal?

I know Knime doesn’t have Nominal as a primitive type, but there is the Domain Calculator node to convert string to nominal… is there a way of making it handle empty values?

[izaychik63]

What about use

Rule Engine — NodePit

with MISSING function?

[Stormdamage]

Not sure what you mean? Putting this into the rule engine does not work:
MISSING $column_name$ => EMPTY

[izaychik63]

Sorry, it could be MISSING $column_name$ => “”
or any other nominal you need.

[mlauber71]

If you want to artificially create missing values you could see if this trick could help.

Create or replace column with NULL values

Hi! These are examples from the Rule engine node: // enter ordered set of rules, e.g.: // $double column name$ > 5.0 => "large" // $string column name$ LIKE "*blue*" => "small and blue" // TRUE => "default outcome" But you do not need any rule. Simply connect it and check Append Column if you want new column. Give it name and new column with missing values will be created Br, Ivan

Discussion also here:

Replace specific row values of a column with NULL values

Sometimes Rule Engine, Variables and Null values can be tricky. I have added an example where you could define your own replacement rules. Age gets replaced by NULL and gender by “NA”. Please note: there is no generic NULL in the Rule engine but ipazin came up with an clever solution (Create or replace column with NULL values) - create a column with just missing and use that you would always need a TRUE rule that covers everything else. If you do not have one you will create missing values yo…

[Stormdamage]

Thanks everyone. The problem is not creating missing values, the problem is converting the existing missing values into a format that the GeneralizedSequentialPatterns (GSP) Weka node will interpret as “empty nominal”.

I have not managed a find a way to achieve this yet, which is frustrating, as the Node is supposed to accept information of varying lengths, which is going to mean some columns have missing values.

For now I have just turned a missing values into a string “-”, but it means they get processed by the Node rather than ignored.

[mlauber71]

It could help if you could provide us with an example with data and the Weka Node and a hint at exactly which point the problem with the missing values exists. Then it may be easier too see what can be done about it.

[Stormdamage]

Thanks, here’s an example workflow. The first column of the table has to be an ID for the GSP algorithm. This column and the sequence columns need to be nominal, so I used the data calculator node to convert them for strings to nominals. However, it seems that empty nominal values are not possible (so i have used “-” here instead). but then this means they get processed by the GSP node instead of being ignored.

GSP.txt (114 Bytes)
GSPTest.knar.knwf (17.3 KB)

[ipazin]

Hi there!

Instead of “-” in Missing Value node just leave empty and you will get empty string which should be ok for your node.

Domain Calculator node scans the data and updates the possible values list and/or the min- and max values of selected columns. It does not convert string column to be nominal data type.

Br,
Ivan