Build an AutoEncoder Knime

To whom it may concern,

I am looking for help/advice i am planning to build an autoencoder to use it to detect fraud thats my project my plan is to build the autoencoder and feed it a dataset with no anomalies so it detects them once it is given a data with anomalies but i need help on how to actually build it ? and what nodes to use and how to start it is it possible i find help i will really appreciate it i am really interested in deep learning and machine learning as me myself i study BSc Computer Science

Welcome to the forum.

There is a fraud detection example on the KNIME Hub. I would start there:

2 Likes

Thank u for this reply i will give it a try this means a lot btw so i can implement keras autoencoders in knime ?? as i can see ive done 1/2 already now i want to do 3 / 4 train the autoencoder and set it up but there are keras nodes where do i get these from ? hmm

The Keras nodes are available from the normal KNIME update site. You’d install it by going to File > Install KNIME Extensions > KNIME Labs Extensions > KNIME Deep Learning - Keras Integration.

Alternately, you can install them via Nodepit.

Or you can open the example workflows and install the nodes when prompted.

1 Like

Okay i will do that tomorrow morning its 2 am for me and i will start testing do you mind if i get your email or any one of your social media its the first guy i meet that has an idea on what i need its my final year project and i am doing fraud detection using AI i wanna show you what i have implemented already which is a churn predication and deployment tree predicator on class and now building an auto encoder i really appreciate your help and hope you give me ur details or a place where i can contact you always for help if needed in knime as it will be all in knime thank you so much again this means a lot you just relieved me from loads of stress if this works for me

@6Vibes Concerning the setup of Keras and KNIME you might want to consult the official guide(s)

If you run into problems or want more control over the setup there is this older thread that might still be relevant

Also H2O offers an autoencoder. There is no KNIME integration but the resulting MOJO files might be able to be imported and used in KNIME - I have not yet set up an example.

There was an example on the forum some time ago also by @menuetto:

One final remark: if this is for an educational assignment my advice would be to make sure you add your own variation and effort to the project.

2 Likes

Im thankful for your response i will defo try setting it all up today so i can proceed i currently installed the deep learning extensions now configuring the conda with it it and trying to finish the installation guide hopefully it works probably

Any idea how to solve this it only did this on my pc but on my laptop it didnt show any errors both versions are the same maybe my pc has windows 2004 updated

Do i need to fix this TensorFlow 2 issue ?? on my pc considering that i will be using the workflows provided by @elsamuel the “MAARIT/EXAMPLES” i tried doing it on my laptop it did not show any errors but i think cuz my laptop is still Windows 10 Home 1909 or idk but not 2004 or 20H2 now i am scared can you provide a solution incase my laptop decides to update or that i can also make it work on my pc

Hey any idea how can i change the threshold level ? and my normalizer keeps giving me this warning "Normalized value is out of bounds Original Value : 3.4 Transformed value : 1.007 Upper bound : 1.0 " @elsamuel