If you are serious on AI/ML, do NOT waste your time on KNIME!

If you are new to KNIME, and want to use KNIME, don’t bother, it is a complete waste of time.
It is very slow, overcomplicated, cannot even solve a basic problem easily.
You are better off to stick to Python, which is what I do.
Python is much simpler, much more efficient.
If do a loop with 10000 rows of data, KNIME is so slow, you are better off even by using excel than KNIME.
It took me 3 full days to learn KNIME to find out the problems. Do NOT trust those who promote KNIME.
I am a geostatistican with more than 20 years experience. I use ML/AI for my work. I tried to find a simpler solution to simplify my work, and make it more efficient. I ended up wasting time.

Could you elaborate on the experiences you had in those three days? Maybe even provide a sample workflow where you had these problems?

KNIME has a dedicated package for Geoinformation tasks (I am not an expert though).

Also it often can make sense to combine KNIME and Python (or Excel) if you have special tasks where there is a good Python package.

6 Likes

First of all - welcome to the forum :slight_smile:

I’m a finance professional with 20 years of experience - I can code in Python / JS etc. as well and still love KNIME.

Also curious to understand better what has triggered such high level of frustration…

10 Likes

Welcome to the forum … and as knimer we would love to know what really made it so such conclusion … which is quite rare who tried this tool.Tool collaborate with every other tool and even for Geospatial knime has dedicate Geospatial extension from CGA-Harvard. Give it a try…and even you feel same… Then You are a diehard PYFan .

5 Likes

Hi @MartynZhu

While it’s understandable that one user had a frustrating experience, that single negative opinion doesn’t reflect the overwhelmingly positive consensus from the broader data science community regarding KNIME. This powerful, open-source platform is widely praised for its intuitive, visual workflow that makes complex data operations accessible to users of all skill levels—from seasoned data scientists to business analysts with no coding experience.

Can you tell us what is your problem?

Br

5 Likes

I can just #2 the other comments here. I’m very interested in your feedback and would love to understand where we could do a better job for interested users like you :slight_smile:

5 Likes

Hi Martyn,

While I respect your experience, I believe your conclusion about KNIME is unfair and overly dismissive.

As someone with a strong programming background myself, I find KNIME to be an excellent complementary tool. It offers great value through its visual workflow design, modular structure, and seamless integration of Java/Python scripts when needed. This hybrid approach makes it both powerful and accessible — especially when building, debugging, and sharing repeatable data workflows.

For example, I use KNIME to run “brute-force” backtests across 6,500+ fund combinations on 400,000+ rows of data and 400 different funds, and the entire process completes in under one hour. That’s far from being inefficient.

Of course, every tool has its place. But dismissing KNIME entirely based on a short experience seems like throwing out the baby with the bathwater. IMHO, once you understand how to leverage its strengths — especially with Python — KNIME can be a serious productivity booster even for experienced data scientists.

Ps: If you need specific help wrt specific cases you are always welcome.

Best regards,
Alpay

1 Like

My reaction may be a bit over the top. Here is one example: when I used loop to deal with 40000 rows of time series data to do rolling window math operations on multiple columns, it takes half an hours, and still did not finish. It is much slower than Excel which takes less than a minute. Eventually I had to use excel to do the calculation, input it into KNIME, and then use python scripts to do the rest of job.
KNIME loop is very inefficient in this case, the only way I could get around it is to use other methods.

2 Likes

I see… what if you could do it without a loop?

I’m not sure about the exact use case (maybe you can share more details), but the Moving Aggregator node may just be what you are after:

If it is not rolling aggregation but averaging then Moving Average node may do it:

Both should be very performant with 40k rows of data.

6 Likes

Hi Martin,

Thanks for the reply.
Aggregator does not do any other math operation other than average, sum, count, min, max. I am trying to do moving window linear regression to obtain the slope, intercept, and variance.

Kind regards

Admittedly I am not an expert in ML by all means - I did a little bit of research on what moving window linear regression is.

If I understand correctly, you

  1. first compute x, x^2 ,x * y, , y^2 for each record.
  2. then aggregate these columns using sum over a moving window
  3. then calculate slope, intercept and variance based on 2) for each row

If the above is correct then:

  1. use either Math Formula nodes or column expressions nodes to calculate the metrics listed under 1) above for each row - this should be really fast
  2. use Moving Aggregation using Sum aggregation method on the new columns created under 1) with a window size of your choice
  3. then use another Math Formula or column expressions node to calculate slope, intercept and variance based on the values generated by Moving Aggregator under 2)

If you can share what you have with an anonymised example data set I’d be happy to give it a crack…

4 Likes

This sounds like you’re struggling with Knime because you’re new to it, not because it’s flawed. I love Knime, but like any processing software it took me time to learn it. In my first couple of weeks I was clumsy and everything felt hard. I’ve been using it for 5 years now and I know it so well. I don’t use python or excel unless I have to.

Hmm, this doesn’t sound right. If I work with 1 million rows Knime gets a bit slow. But 10,000 rows is very fast. Not sure why it is slow for you, but this is not typical Knime performance.

1 Like

Hi everyone,

I’ve added an example here—a randomly generated two-column time series dataset that allows us to calculate regression statistics using a rolling window.

You’ll find both the data and the KNIME flow attached.
It’s not instant, but certainly not an hour either—it took around 2 minutes for me.

Please feel free to reach out if you need help. Alternatively, the best way for us to assist you is if you could provide an example dataset.

Best,
Alpay

KNIME_RollingWindow_40000Rows.xlsx (1.7 MB)
Knime Flow.knwf (17.2 KB)

3 Likes

Hi Alpayzeybek,
Thanks for the reply. I have downloaded your work flow, and will try it.
I was learning KNIME using ChatGPT’s instructions. It was a painfully slow process on my laptop computer, and was frustrating. I hope that the way I learned was wrong.
Kind regards
Martyn

3 Likes

Hi Martyn,

Thanks for your message.

I totally understand—trying to learn KNIME just through ChatGPT can feel a bit overwhelming, especially when you hit topics like loops, time series, or performance tuning. You’re definitely not alone in feeling that frustration. IMHO chatgpt can still not replace knime forum, yet :slight_smile:

The good news is, once you get past those early bumps, KNIME really opens up. It’s a powerful tool—especially when you start mixing its visual workflows with a bit of scripting or clever workarounds.

If you ever feel stuck or just want to sanity-check something, don’t hesitate to reach out.

Kind regards,
Alpay

2 Likes

Congrats! That’s a great start to get attention by many KNIMErs!

If you want to learn KNIME I highly recommend the free training courses https://www.knime.com/learning or check youtube channel https://www.youtube.com/@KNIMETV . It’s really worth to spend some time there.

In general are row by row loops pretty slow no matter if programmed in KNIME or Python. So vectorization is the key! In Python you can use numpy or pandas operations, in KNIME operations in specific nodes like “moving aggregation” are pretty fast and sufficient for the most applications.
But why not use both? KNIME offers Python Script Nodes.

6 Likes

It is quite well known that these wonderful new AI tools are giving a quick start in the beginning. When drilling down to details you will quite often see that details are not (fully) correct and this might stop the own learning curve.
So I would see this as another good example that shows that you can’t (yet) trust what AI is telling.
It’s still worth to use individuals’ knowledge - e.g. in this forum - to go the steps to become a master in KNIME. And don’t hesitate to ask here either to check the answers of AI or just ask for other users’ knowledge and experience.

KNIME is a powerful tool which will can support you. but this power comes with a certain complexity (e.g the huge amount of nodes) or different approaches to solve a challenge. So it’s always worth to share the own knowledge and ask for potential other approaches for a challenge.

Hopefully you will continue to use KNIME despite this thread’s titel.
Maybe you can also consider to adapt this threat’s title.

4 Likes

Oh, I get you—learning KNIME through ChatGPT isn’t always ideal. A lot of times the answers aren’t quite right. Honestly, it’s much better to sign in to KNIME Hub and try out K-AI, their own AI assistant—that one’s way more reliable.

I used to think the same about KNIME. Python felt like the smarter choice, especially when loops or try/catch logic are so straightforward. I kept asking myself, “Why am I using KNIME?”

But that changed. Now, whenever I hit a roadblock, I just drop in a Python Script node inside KNIME and keep going—no time wasted.

What really got me is this: with Python, I’d sometimes look back on my old work and feel lost, trying to figure out what I did in each step. But KNIME makes documentation and workflow tracking super easy. I can open a project months later and still understand everything I did.

Welcome to the KNIME community, Martyn!

6 Likes