#metabrainz

/

      • shivam-kapila
        ruaok: But I dont like that :/
      • 2020-06-29 18154, 2020

      • ruaok
        dont like which one?
      • 2020-06-29 18107, 2020

      • shivam-kapila
        current heading in stats section
      • 2020-06-29 18111, 2020

      • ruaok
        agreed.
      • 2020-06-29 18122, 2020

      • shivam-kapila
        I don t like keeping actionable items in headings
      • 2020-06-29 18138, 2020

      • ruaok nods
      • 2020-06-29 18149, 2020

      • shivam-kapila
        Mr_Monkey: rescue plz
      • 2020-06-29 18103, 2020

      • ruaok
        lol
      • 2020-06-29 18110, 2020

      • Mr_Monkey
        What's up?
      • 2020-06-29 18115, 2020

      • jmp_music
        alastairp: this is gaia's best model params for the `danceability`: https://usercontent.irccloud-cdn.com/file/KUDpvcc…
      • 2020-06-29 18138, 2020

      • shivam-kapila
        I asked my UI friends too, they are also confused
      • 2020-06-29 18149, 2020

      • shivam-kapila
        Mr_Monkey: The sidebar story
      • 2020-06-29 18100, 2020

      • shivam-kapila
      • 2020-06-29 18103, 2020

      • jmp_music
        alastairp: and this is a single train/test split evaluation with the same params in sklearn: https://usercontent.irccloud-cdn.com/file/mliWnFW…
      • 2020-06-29 18117, 2020

      • Mr_Monkey
        I haven't looked at them yet, let me see
      • 2020-06-29 18151, 2020

      • alastairp
        jmp_music: what features is the sklearn version using?
      • 2020-06-29 18109, 2020

      • shivam-kapila
        ruaok: Meanwhile can yopu look at Desktop Concept 2 -- how do the cards go listen line
      • 2020-06-29 18112, 2020

      • shivam-kapila
        list*
      • 2020-06-29 18143, 2020

      • jmp_music
        alastairp: well, I follow a little but different preprocessing procedure compared to gaia, till now. I use all the numerical features, and I enumerated the categorical ones too. After that, I let PCA to decide which features are highly correlated and cover the 95% of the dataset's variance.
      • 2020-06-29 18154, 2020

      • shivam-kapila say Mr_Monkey going to DC2
      • 2020-06-29 18157, 2020

      • shivam-kapila
        saw*
      • 2020-06-29 18117, 2020

      • ruaok
        sorry, what should I be looking at in DC2?
      • 2020-06-29 18136, 2020

      • shivam-kapila
        The list of listens
      • 2020-06-29 18146, 2020

      • shivam-kapila
        Each listen is in a card
      • 2020-06-29 18156, 2020

      • jmp_music
        alastairp: gaia uses different procedure. It checks the mfcc, it deletes some of the numerical too, but I experimented with all the features to see what the results will be
      • 2020-06-29 18157, 2020

      • shivam-kapila
        Rather than separated by horizontal line
      • 2020-06-29 18102, 2020

      • ruaok
        looks fine to m.
      • 2020-06-29 18154, 2020

      • Mr_Monkey
        As far as I'm concerned, I think the two best options are side bar 2 and combined nav. I find MP 3 to 6 confusing one way or another.
      • 2020-06-29 18103, 2020

      • jmp_music
        alastairp: That's what i saw in the config file. gaia checks and tests all the possible methods
      • 2020-06-29 18119, 2020

      • jmp_music
        both in scaling, and the feature selection
      • 2020-06-29 18136, 2020

      • shivam-kapila
        Mr_Monkey: Thats the only problem. Confusion
      • 2020-06-29 18111, 2020

      • Mr_Monkey
        I think the options that is most unambiguous, easiest to implement and saves screen real-estate (scrolls out of view) is the combined nav
      • 2020-06-29 18113, 2020

      • Mr_Monkey
        Or, to be more precise, not combining the bootstrap hamburger nav and the side bar nav, but on mobile always show the side nav on top, not collapsible.
      • 2020-06-29 18127, 2020

      • jmp_music
        alastairp: That's why I decided to not adopt a static method, but let PCA decide dynamically the best features at each training phase, based on the nature of the dataset. I want now to implement gaia's static feature selection too, to compare with my previous skleran implementation.
      • 2020-06-29 18143, 2020

      • alastairp
        jmp_music: great. as our initial goal is to completely replicate the gaia results, the static feature selection looks like a good next step
      • 2020-06-29 18104, 2020

      • shivam-kapila
        Mr_Monkey: I got you
      • 2020-06-29 18132, 2020

      • ruaok
        I agree that in mobile, the sidebar doesn't work -- the horizonal space is too precious.
      • 2020-06-29 18108, 2020

      • jmp_music
        alastairp: cool! When I reach that step, I would like to ask you which type of scaling gaia uses for normalization, and guassianization on data scaling. But first let me check the cpp code
      • 2020-06-29 18127, 2020

      • jmp_music
        alastairp: I found the code inside the algorithms directory
      • 2020-06-29 18117, 2020

      • jmp_music
        Additionally, I saw that gaia triggers PCA only when the features selection is above 80!
      • 2020-06-29 18134, 2020

      • jmp_music
        that's in pca.cpp file
      • 2020-06-29 18108, 2020

      • alastairp
        oh cool, I didn't even know that gaia does pca!
      • 2020-06-29 18121, 2020

      • shivam-kapila
        Mr_Monkey: Can you look at mine one
      • 2020-06-29 18128, 2020

      • alastairp
        for the normalization and gaussianization it will be best if you look at the code
      • 2020-06-29 18125, 2020

      • Mr_Monkey
        shivam-kapila: So something like that can work I think
      • 2020-06-29 18158, 2020

      • shivam-kapila
        looks like this on mobile
      • 2020-06-29 18159, 2020

      • shivam-kapila https://www.figma.com/proto/PSXpqDEQ7Melwj14hBLZaj/ListenBrainz-Dashboard-(Experimentation)?node-id=396%3A427&scaling=scale-down
      • 2020-06-29 18105, 2020

      • shivam-kapila
        ruaok: Mr_Monkey ^^
      • 2020-06-29 18146, 2020

      • Mr_Monkey
        Not sure you need the white tab to have space above it, and considering you already have the section title (recent listens) you can avoid repeating it just underneath
      • 2020-06-29 18154, 2020

      • ruaok
        that whole line is a bit odd to me -- a lot of space dedicated to just showing a name.
      • 2020-06-29 18127, 2020

      • ruaok
        but overall this design is starting to grow on me.
      • 2020-06-29 18144, 2020

      • shivam-kapila
        ruaok: how to indicate user_name then?
      • 2020-06-29 18125, 2020

      • shivam-kapila
        Mr_Monkey: that space is something I like 😬
      • 2020-06-29 18126, 2020

      • ruaok
        thats just it, I don't have better suggestion.
      • 2020-06-29 18101, 2020

      • ruaok
        also, the word Beta take up extra space. maybe time to get rid of it?
      • 2020-06-29 18128, 2020

      • shivam-kapila
        we may move that beta with timescale migration
      • 2020-06-29 18136, 2020

      • shivam-kapila
        remove*
      • 2020-06-29 18159, 2020

      • ruaok
        yeah, when we have rolling releases on a frequent basis, what does beta even mean?
      • 2020-06-29 18128, 2020

      • shivam-kapila
        beta == son in hindi😛
      • 2020-06-29 18154, 2020

      • shivam-kapila
        Mr_Monkey: Waana give some final touches and then we code it out?
      • 2020-06-29 18127, 2020

      • shivam-kapila
        ruaok: you too please
      • 2020-06-29 18157, 2020

      • jmp_music has quit
      • 2020-06-29 18103, 2020

      • ruaok
        final comments?
      • 2020-06-29 18120, 2020

      • shivam-kapila
        touches comments all work
      • 2020-06-29 18126, 2020

      • ruaok
        I dont think I have any. LGTM.
      • 2020-06-29 18153, 2020

      • shivam-kapila
        One question -- Do we keep those icons in sidebar in desktop mode?
      • 2020-06-29 18106, 2020

      • jmp_music joined the channel
      • 2020-06-29 18112, 2020

      • Mr_Monkey
        Yes
      • 2020-06-29 18124, 2020

      • Mr_Monkey
        If we use them in one view we use them in the other, for consistency
      • 2020-06-29 18153, 2020

      • Mr_Monkey
        Probably right-aligned in desktop side bar
      • 2020-06-29 18137, 2020

      • shivam-kapila
        DC1 looks cool
      • 2020-06-29 18147, 2020

      • shivam-kapila
        Not sure if its to be orange though
      • 2020-06-29 18157, 2020

      • Mr_Monkey
        shivam-kapila: Either of those tab options would work for me. I see where the design with the rounded tab comes from, but I think in this view it's distracting (no used to see tabs quite like that)
      • 2020-06-29 18110, 2020

      • Mr_Monkey
        Let me try sometihng quick
      • 2020-06-29 18141, 2020

      • iliekcomputers
        ishaanshah: hey, I'm around
      • 2020-06-29 18157, 2020

      • ishaanshah
        iliekcomputers: give me two mins
      • 2020-06-29 18117, 2020

      • shivam-kapila
        Mr_Monkey: sure sure
      • 2020-06-29 18122, 2020

      • iliekcomputers
        Sure
      • 2020-06-29 18102, 2020

      • ishaanshah
        iliekcomputers: Hi!
      • 2020-06-29 18112, 2020

      • jmp_music
      • 2020-06-29 18130, 2020

      • ishaanshah
        I continued working on the frontend today
      • 2020-06-29 18151, 2020

      • jmp_music
        alastairp: I think that with `descdim` means descriptors dimensions
      • 2020-06-29 18103, 2020

      • ishaanshah
        The line graph wasn't working for monthly listening activity
      • 2020-06-29 18113, 2020

      • iliekcomputers
        Wasn't working?
      • 2020-06-29 18118, 2020

      • alastairp
        jmp_music: right. however, also check this: https://github.com/MTG/gaia/blob/ed433ed3f1fa3cee…
      • 2020-06-29 18130, 2020

      • alastairp
        I _think_ that we only use classifier type 'svm'
      • 2020-06-29 18131, 2020

      • ishaanshah
        not working in the sense not useful
      • 2020-06-29 18143, 2020

      • iliekcomputers
        Ah, what was the issue?
      • 2020-06-29 18147, 2020

      • ishaanshah
        Too much data
      • 2020-06-29 18155, 2020

      • alastairp
        and we only do rca/pca in 'rcasvm', 'mixedsegments', and 'mixed'
      • 2020-06-29 18156, 2020

      • ishaanshah
        nothing could be made out clearly
      • 2020-06-29 18100, 2020

      • iliekcomputers
        Hmmm
      • 2020-06-29 18108, 2020

      • iliekcomputers
        Do you have a screenshot maybe?
      • 2020-06-29 18114, 2020

      • ishaanshah
        So I switched to Bar graph as you said
      • 2020-06-29 18124, 2020

      • iliekcomputers
        That sounds good to me!
      • 2020-06-29 18157, 2020

      • ishaanshah
        > Do you have a screenshot maybe?
      • 2020-06-29 18157, 2020

      • ishaanshah
        I don't, but I can use git to revert
      • 2020-06-29 18100, 2020

      • jmp_music
        alastairp: So it doesn't use pca and rca when it calls the SVM for training
      • 2020-06-29 18100, 2020

      • Mr_Monkey
        shivam-kapila: I can't figure out how to do this in Figma: https://cdn-images-1.medium.com/max/1200/1*MACsg7…
      • 2020-06-29 18115, 2020

      • alastairp
        jmp_music: I believe so
      • 2020-06-29 18121, 2020

      • ishaanshah
        But the bar graphs make more sense functionally
      • 2020-06-29 18124, 2020

      • iliekcomputers
        ishaanshah: no need really, I was just curious
      • 2020-06-29 18128, 2020

      • Mr_Monkey
        rounded top to the tab with a border on the sides and top but not on the bottom.
      • 2020-06-29 18143, 2020

      • iliekcomputers
        Agree about the bar graph
      • 2020-06-29 18143, 2020

      • Mr_Monkey
        (if you want the tab desing to be visible without having white on grey background)
      • 2020-06-29 18155, 2020

      • ishaanshah
      • 2020-06-29 18108, 2020

      • Mr_Monkey
      • 2020-06-29 18121, 2020

      • iliekcomputers
        That looks great!
      • 2020-06-29 18134, 2020

      • jmp_music
        alastairp: Thus, we are going to replicate only the static methods for start
      • 2020-06-29 18143, 2020

      • ishaanshah
        I wanted to talk about some high level stuff for listening activity
      • 2020-06-29 18152, 2020

      • iliekcomputers
        Sure, shoot!
      • 2020-06-29 18125, 2020

      • alastairp
        jmp_music: perfect. Have a look at see if you can find the details for the enumeration, normalization and gaussianization
      • 2020-06-29 18127, 2020

      • Mr_Monkey
        shivam-kapila: well, duh! I'm stoopid. I should have pointed to the tabs we currently have in LB :p
      • 2020-06-29 18130, 2020

      • shivam-kapila
        Mr_Monkey: give me some time
      • 2020-06-29 18134, 2020

      • alastairp
        tell me if you have any problems
      • 2020-06-29 18141, 2020

      • ishaanshah
        The main issue is the horizontal space
      • 2020-06-29 18144, 2020

      • Mr_Monkey
        No need to mock it up, really
      • 2020-06-29 18103, 2020

      • ishaanshah
      • 2020-06-29 18117, 2020

      • ishaanshah
        We cant fit the data for last month as well as this month
      • 2020-06-29 18126, 2020

      • ishaanshah
        especially on mobile phones
      • 2020-06-29 18130, 2020

      • iliekcomputers
        The card with listens per day etc
      • 2020-06-29 18106, 2020

      • iliekcomputers
        Could we put that on a horizontal line and then use the newly freed up space for the graph
      • 2020-06-29 18111, 2020

      • ishaanshah
      • 2020-06-29 18119, 2020

      • jmp_music
        alastairp: Cool! The enumeration I do in sklearn, is exactly applied to the same features as gaia does. I'll ask you for the rest ones if I find any difficulties. Thanks!
      • 2020-06-29 18119, 2020

      • ishaanshah
        ignore the overlap below
      • 2020-06-29 18125, 2020

      • ishaanshah
        gonna fix that tomorrow
      • 2020-06-29 18105, 2020

      • ishaanshah
        > Could we put that on a horizontal line and then use the newly freed up space for the graph
      • 2020-06-29 18105, 2020

      • ishaanshah
        We do that on mobile, still doesn't fit as shown in the screenshot above
      • 2020-06-29 18101, 2020

      • ishaanshah
        So there are two things we can do
      • 2020-06-29 18103, 2020

      • iliekcomputers
        Hmm
      • 2020-06-29 18115, 2020

      • ishaanshah
        One is convert it to a stacked bar graph
      • 2020-06-29 18126, 2020

      • ishaanshah
        Or skip last month for monthly stats
      • 2020-06-29 18143, 2020

      • ishaanshah
        I am inclined towards the second
      • 2020-06-29 18146, 2020

      • ishaanshah
        option
      • 2020-06-29 18147, 2020

      • iliekcomputers
        We could also make the graph vertical
      • 2020-06-29 18110, 2020

      • ishaanshah
        That would eat up too much space
      • 2020-06-29 18121, 2020

      • ishaanshah
        We also have other stats below
      • 2020-06-29 18159, 2020

      • iliekcomputers
        Skip last month for monthly stats on mobile then
      • 2020-06-29 18125, 2020

      • ishaanshah
        I think we should skip it for desktop too
      • 2020-06-29 18141, 2020

      • iliekcomputers
        Why?
      • 2020-06-29 18157, 2020

      • ishaanshah
        the bars already look too thin IMO
      • 2020-06-29 18108, 2020

      • iliekcomputers
        If we add the space from the card, they won't.
      • 2020-06-29 18129, 2020

      • iliekcomputers
        The data is interesting, I want to see it 🤷🏽‍♂️
      • 2020-06-29 18143, 2020

      • ishaanshah
        Hmm, that makes sense
      • 2020-06-29 18104, 2020

      • ishaanshah
        So essentially it will two row layout for both mobile and desktop right?
      • 2020-06-29 18136, 2020

      • iliekcomputers
        Right, and the first row (listens per day etc) doesn't need to be very tall.
      • 2020-06-29 18114, 2020

      • yvanzo
        bitmap, reosarevok: Updating mb.o website/web service containers.
      • 2020-06-29 18116, 2020

      • ishaanshah
        Yep makse sense
      • 2020-06-29 18123, 2020

      • ishaanshah
        makes*
      • 2020-06-29 18145, 2020

      • iliekcomputers
        👍🏽