How to interpret the confusion matrix especially when it is having question marks...


With help of this amazing forum I did eventually manage to ran my deep neural network node.
But I seriously have doubts whether confusion matrix shown by it is accurate or not. And I am asking myself whether this model have learned something or not.
I guess what I am asking is how to interpret the results of confusion matrix in Score node, which I am finding little different than other statistical software I am used to.

But I am happy that I was able to run my first deep neural network just within a week. I would like to read about this epoch and learning curves. Is there any manual about deep learning extension about knime which would explain the details of of the variables one needs to choose before running the network.
Also I wanted to ask how many dense layers one should add is there any optimum number.
I guess my first exercise has left me with many more questions.
Thanks to you guys I am still curious in learning about it more.

With regards

Hello Akshaykumar,

the statistics you posted are not the confusion matrix, in order to show it, you can either open the view of the scorer node (press F10 or select it in the context menu), or the first outport.

The missing values are for the most part expected e.g. the Accuracy is not defined on a per class basis but for the overall table. What this does tell you is that the model is right in 64.3% of the rows but also that it never predicts class 1, which is why Precision and F-measure are missing for this class.

We don’t have anything like a deep learning manual but we are using the same algorithms and approaches that are used in many other deep learning frameworks (we actually just integrate them into KNIME).
So any kind of deep learning course or tutorial you read on the web will give you insights that you can apply to your KNIME workflow, especially as it seems that you are just starting out with deep learning and machine learning in general.
I would also recommend learning a bit more about machine learning in general, e.g. what kind of metrics can be used in which applications and how to evaluate your models properly.
For this, you can also have a look at webinars and videos we posted on our YouTube channel.
I can also recommend checking out Udacity for free courses on machine learning and deep learning, a while back there was one from Google that I found quite helpful and I guess it should still be available.