Please share with the community what you think needs improvement with RapidMiner.
What are its weaknesses? What would you like to see changed in a future version?
I have the deep learning models on my laptop but it doesn't work very well. I think that they should make deep learning models easier. They are using deep learning models today for image processing and language processing.
I think it is a fairly straightforward interface generally. It is an easy-to-use solution compared to SAS Enterprise Miner, for example. On the other hand, compared to some other products, maybe the UI could be enhanced. The visual interface could have something like the-drag-and-drop features which Alteryx already supports. Some of those additional features can make RapidMiner a better tool and maybe more competitive or advanced.
When I started using RapidMiner, I found it difficult to get it to read the metadata. I wanted to use, for example, a pivot table, and it did not have the variable or the attribute names in it. There were no values. It took a long while to figure out how to do that, although it tends to do it automatically nowadays. RapidMiner is not utterly intuitive for beginners. Sometimes people have trouble distinguishing between a file in their own file system and a repository entry, and they cannot find their data. This is an area where this solution could be improved. It would be helpful to have some tutorials on communicating with Python. I found it a bit difficult at times to figure out which particular variable, or attribute, is going where in Python. It is probably a simple thing to do but I haven't mastered it yet. I'd like them to do a video on that. There are a large number of videos that are usually well-produced, but I don't think that they have one on that. Essentially, I would like to see how to communicate from RapidMiner to Python and from Python to RapidMiner. One of the things I do a lot of is looking at questionnaires where people have used Likert-type scales. I don't recommend Likert-type scales, but if they're properly produced, which is a lot of hard work and it's not usually done, they're really powerful and you can do things like normalizing holes on the Likert scale. That's not the same as normalizing your data in RapidMiner. So, I would want to get results with these Likert scales, pass it through RapidMiner, do a normalization and pass back both the raw scores and the normalized scores and put in some rules, which will say if it's high on the raw score and on the normalized score and low on the standard deviation, then you can trust it.
The biggest problem, not from a platform process, but from an avoidance process, is when you work in a heavily regulated environment, like banking and finance. Whenever you make a decision or there is an output, you need to bill it as an avoidance to the investigator or to the bank audit team. If you made decisions within this machine learning model, you need to explain why you did so. It would better if you could explain your decision in terms of delivery. However, this is an issue with all ML platforms. Many companies are working heavily in this area to help figure out how to make it more explainable to the business team or the regulator.
I think it's a great product but confusing in some way with regard to the user interface and integration with other tools. An improvement would be the addition of some buttons which would be useful because I'm sometimes unsure why I need to use something or what is its purpose. I would say the same goes for additional features, the addition of buttons would be helpful. The product is better than other software that I use.
RapidMiner would be improved with the inclusion of more machine learning algorithms for generating time-series forecasting models.
I would like to see all users have access to all of the deep learning models, and that they can be used easily. RapidMiner loads very slowly, which is something that should be improved.
The price of this solution should be improved.