Jovan Veljanoski

> @JovanVeljanoski any opinions on this? How should we attach this, or do you like my code proposal? Still thinking about it.. i want to do some tests but busy......

I like the proposal of @maartenbreddels above. The one correction/suggestion I would make is this ```python df['person'] = df.struct.merge(['name', 'age']) df['person'] = df.struct.merge({'name':'Name', 'age':'Age'}) ``` Although I have to say...

Very hard to tell without a code example that we can reproduce. Also there were bunch of questions in the form that you didn't answer.

I appreciate the example and insight into what you are doing, but this is not something I can copy paste and run & debug. Can you provide some (fake) data...

In both functions you are mixing vaex and non-vaex code (numpy/pandas). That might cause memory spikes. It might be a good idea to add bunch of print statements and time...

How many unique values do you have in `tm_cid` and `tm_mid`?

There has been few releases since this thread was active. @ashsharma96 can you see if things have improved for your usecase in the latest version?

@maartenbreddels (if understand this correctly..) What would happen in this case. Say you have a original column without N/A values. You do some computation using that column, and overwrite it...

Hi, Interesting PR. Here are my thoughts: 1. Regarding `df.to_numpy()` proposal: Something very similar already exists actually, it is called `df.to_arrays()`, where by default you get the data in the...

Another option for the PCA issue would be to not modify the PCA at all, but also simply not support this `features_` property in this case.