Online real-time scoring


Is it possible create a predictive model like the one in the sentiment analysis and deploy the model on a server and then score new data online and real-time?

Since the  parameters are many more (when taking into account word stems or ngrams) in text mining, each new data (a new comment on a movie, e.g.) to be predicted should be pre-processed with the data used to train the model to prevent any difference of data structures of the trained model and the new data to be predicted. We know this causes an error like ""WARN Decision Tree Predictor Learning column "Prediction (order)" not found in input data to be predicted" as in this post.

However, this situation requires a huge workflow  and a long pre-processing (at least). Even on Knime in my local computer, this process takes around several minutes.  All of these in mind, is it really possible and feasible to real-time-score new data in text-mining cases on server ?



Hi Bora,

we will work on a REST api for the KNIME Server in 2015. Workflows and models will be preloaded on the KNIME Server and kept in memory. Single (or multiple) datapoints (documents) can be send to the REST api. They will be processed according to the workflow and the model will do the prediction. This setup goes into the direction of real time scoring / prediction.

Of course you need to ensure that the feature space of the data on which the model will be applied is equal to the feature space of the training set. This is a little bit tricky for documents but doable. Additional columns have to be filtered and missing columns added with values of 0.

Cheers, Kilian

Thanks, Killian.

I am curious about the price range of this service. Is it possible to speculate about the pricing at this moment?




Hi Bora,

the REST API will be part of the regular KNIME Server:

Cheers, Kilian

Hi Kilian,

The current SOAP API does not accept datapoints?

So I am guessing the REST API will be released in KNIME 2.12, which is next month.



Hi Kenny,

no the SOAP API does not accept data as input. The REST API of the server will be release with the new server version. For more details see the blog article:

Cheers, Kilian