Significant Improvements

Akmal · June 23

I work as a data scientist manager at Pfizer. I have some feedback for DSS.

1. Does not return to the exact location in the Flow

- I have a dataset in zoneA and it is shared to another zoneB.

- I explore the dataset in zoneB

- When I come back to the Flow, the flow focuses on the dataset in zoneA instead.

2. Cannot compare different experiments in the Experiment Tracking

In other tools like MLflow, you can easily do it

3. Cannot directly go to the experiment detail page from the visualization

4. Cannot save the filtered column view

- I am in the Explore view page of a dataset.

- I choose only to display column 1

- I go to Flow

- I go back to the Explore view of the same dataset, it shows all the columns again

5. The quality of the downloaded chart is too poor, worse than a direct screenshot.

I want a good quality image to show it to important business stakeholders in a Power Point presentation.

6. Has to type everything in Python recipe because there is no any IDE support, e.g. autocomplete

and Code Studio is really slow...

7. Impossible to append output dataset directly into input dataset

Currently you have to:

1. Create an output dataset

2. use the stack recipe to append it to the input dataset

8. "Why I have to manage the sync of python recipe and python notebook myself?"

In the python notebook, there are both "Save back to recipe" and "Save" buttons. But it is really confusing what is the difference.

The problem is that both Python recipe and Python notebook are writing to the same dataset.

Because I have to manage both the python recipe and notebook, sometimes I just accidentally override the dataset..

9. The concept of charts and insights is very confusing

During the experiment phase, I need to test out different charts to see which one to use.

Currently if I want to change a chart, you need to remove everything and create a chart and create insights and put it in the dashboard.

Why not directly change the chart and the dashboard updates?

10. I wanted to show upper, lower (confidence interval) but you cannot find the right chart type

- I ended up exporting the chart and putting it in a painting software to do it..

Hope you guys can take into account my feedback!

Operating system used: Cloud Based Solution

Han_Han · June 23

Hi @Akmal,

Thanks for taking the time to give us feedback and it is great to hear back from the users who is actively using the DSS!

I will look into your feedback and if i have some follow up questions. I will reach out to you. Do not hesitate if you have any question too. My email address is: han.han@dataiku.com

Best,

Han

Turribeach · June 23

https://community.dataiku.com/discussion/45082/significant-improvements

For Product Enhancements please use the Product Ideas forum. Having said that your post has too many points a lot of which lack clear detail or have possible solutions so it's probably best to discuss it here first.

Check that the "Remember your position in the flow" option is enabled in your Dataiku user's profile. I can't reproduce your issue with that setting.
NC
NC
I can't reproduce your issue. Can you make a video of the issue happening?
How are you downloading the chart? Explain exactly the steps you take as there are several ways to achieve this. Post evidence of the quality issues.
Visual Studio Code in Code Studios is a very good IDE. It's not slow in itself and works really well. Perhaps you have created a Visual Studio Code in Code Studios template with too many code environments? This makes the Visual Studio Code in Code Studios template too big and takes time to load. How big is your Visual Studio Code in Code Studios template image? Describe your slowness problem in detail. What is sloiw? How is it slow? Be exact about your statement, slow is not really precise at all.
Circular flows are allowed since v13.
"The problem is that both Python recipe and Python notebook are writing to the same dataset". This is not a problem, it's by design. You have the two options to allow you to debug your recipes and work with them in a more interactive environment like Jupyter Notebooks. But you don't have to have both. You can leave recipes as recipes and notebooks as notebooks if you prefer. It's your choice. If you follow simple guidance to save back to the recipe you will not have any issues. Of course the code writes to the dataset, that's what supposed to do. Not sure why you expect it to behave differently in a notebook. The save / save to recipe is not confusing. While you are debugging you can save the notebook but you may not want to save back to the recipe which is what Dataiku executes. Once you finished debugging you can either save to recipe or ignore and leave without changes. Your choice.
Chart ⇒ Insight ⇒ Dashboard. That's a pretty simple workflow. Don't publish to Insights until you know what you want then you won't have to redoit.
Dataiku's charting engine it's clearly not going to be the top choice but it does a good overall job. But the beauty is that you are not forced to just Dataiku charts and can also use other Python visualisation frameworks like Plot.ly or Dash WebApp or Bokeh WebApp.