KNIME if-ing easily

There are lots of if and case nodes in KNIME but too many to remember so I use Java If and write my one most of times.

Very easy right? This allows to do any if-ing.

Transfer Learn like thing using freeze layer node

It's a bit of nodes but Keras freeze layer in the top middle is everything here. I retrained last layer like this.

Is this meaningful???
Maybe not, I mean it didn't make the score any good nor bad.

Looking at this KNIME blog post

https://www.knime.com/blog/transfer-learning-made-easy-with-deep-learning-keras-integration

Should I add layer after training by original model? and is it enough data anyway? There are so many things I can play around with this now but I should do with real life data. Cool play around awaits!!

Hierarchical Clustering

I realized I didn't do this before but do tend to do this to see the tree. So here is the simple workflow. So Simple and GUI by KNIME.

and the result

Yes, too many data points. But wanted to see how much time it takes to calculate this.
6000 datapoints and 1024 ECFP. I waited for 2 hours, and didn't finish. Went to sleep and it was done. haha, nice experiment.

slept funny and my neck to shoulder hurts... couldn't go to golf practice... damn...

Favourite string modification nodes

I occasionaly come across a situation wanting to change string by rules. So rule engine is what I use like 0 => Low, 1=> Mid, 2=> High and so on.

Another one is column expression where I can modify in one go. Not a main nodes but tend to use them a lot!

ディープラーニング検定があるってばよ

本屋さんでこんな本を発見

https://www.amazon.co.jp/dp/4798157554/ref=cm_sw_em_r_mt_dp_U_gb2ADbKRSEFBN

ディープラーニング検定なんてあるんだ…気になるので本を買ってみました。

内容はディープラーニングの歴史、手法、応用分野など。ディープラーニングの知識がテストされる模様。

それよりもpythoningとか数学のところとか知りたいのになーと思ったところ、E検定なるものがそれっぽいものみたい。ただE検定の勉強本は無い？Pythonでディープラーニングとかならいくらでも本は出てるので、それで勉強するのかな？

Added Descriptors and Options for previous Neural Network? Workflow

 

in previous post I made Neural Network like workflow. I was wondering how to add descriptors as an input, so I tried it too. It was VERY easy... I thought I would have more trouble doing this.

So, I added descriptors, normalizer, collect column nodes (on bottom left of the workflow). Other than this, the input needed to be adjusted to 

the conversion needs to be "double", that's all about it.

Also, I was wondering where you configure parameters like # of epochs, learning rate, early stoppings, but it was all in the Keras network learner node's options.

GUI to allow you do this is quite awesome! 

hateknime.hatenablog.com

KNIME 4.0.1 released

KNIME 4.0.1 was released few days ago.

https://www.knime.com/changelog-v40

hmmm... lots of bug fixes as I would imagine for 4.0.1. 

I like this enhancement though

• AP-12119: Copy&Paste of workflow from Hub
• AP-12001: UI improvements to the metadata editor

but still more of a bug fixes that we've been 

• AP-12484: XGBoost Predictor can't deal with different column orders

like this XGBoost bug fix is a good fix for me.

I also get lots of Java heap space dumps that I would like to see not a freeze in the program but some work around. Although changing the KNIME setting knime.ini could help too.