Cool project, but missed opportunity to name the arbitrary dimensions Y and C...
lovestory 25 days ago [-]
My dumb ass was trying to figure out what each dimension meant
tptacek 25 days ago [-]
That doesn't make you dumb; there is no intuitive meaning for the axes chosen; you can think of them, roughly, as statistically chosen to maximize clustering.
bravura 25 days ago [-]
Statistically chosen to maximize *some particular loss measure, which in this case might be the t-SNE or UMAP criterion, and is computed only globally and not for different filters.
tptacek 25 days ago [-]
Right (I mean, I'm saying "right" but really I should just say "I'm taking your word for it"), but even more fundamentally this is dimensionality reduction from an OpenAI embedding vector, which seems almost like the asymptotic limit of inscrutability.
alex-knyaz 25 days ago [-]
same
Bilal_io 25 days ago [-]
OP made the change
yoouareperfect 25 days ago [-]
haha awesome, shipped!
ProofHouse 25 days ago [-]
I figure why not plot them with an X and Y (Y,C) of some sort
paxys 25 days ago [-]
There's no need to include an X & Y axis, labels and gridlines if they all have no meaning. A simple cluster diagram is enough.
ascorbic 25 days ago [-]
I agree it would be less confusing if they weren't there. I'm sure I'm not alone in spending some time trying to work out what the axes were.
mbostleman 24 days ago [-]
That’s why I’m here
rl_for_energy 25 days ago [-]
It’d be nice to just see the name of the company on click instead of going to the website (I’m on mobile). Trying to find our company
HeyTomesei 24 days ago [-]
Looks nice, but I'm lost. What do the colors represent? What do the axes #s represent?
kurayashi 23 days ago [-]
The colours represent the categories in the filters. Sadly they don’t show which category is which colour.
default_ 23 days ago [-]
Improvement Proposition: Could you plot the investment size on the C-axis and the number of people working for the company on the Y-axis? The chart should be improved; otherwise, it lacks meaning.
rrr_oh_man 25 days ago [-]
Cool concept! What are the X and Y axes?
Oh, and your website has an unchanged Wordpress favicon...
tptacek 25 days ago [-]
They're semi-arbitrary, dimensionally reduced from OpenAI embedding vectors.
crush_robo_1536 24 days ago [-]
Love this! It'd be interesting if some builds this but adds more dimensions (similar to Company status) to it that you can query or group by. For example, if I look at S21 and W21 batches, then it'd be nice to know things like -
1. How many of these companies made it to series A, series B, etc
2. How many of these companies have > x employees (where x can be 5, 10, 20, etc)
3. How many of these companies had a founder that moved on to something else
This does require a lot more intelligent data scraping or manual data collection though.
iceman_w 21 days ago [-]
I've been scraping YC data week over week to track things like changes in founder, pivots in the idea, company shutting down, etc. You can check it out here https://pivots.fyi/
zild3d 25 days ago [-]
fun, though I also got stuck on what the Y and C axes represent initially. IMO just hide the axes altogether, since the goal is just some visual clustering/similarity
skeeter2020 25 days ago [-]
Maybe I'm slow, but clustering on what dimension? The lack of axes and labeling makes it pretty confusing to me, but I'm a dinosaur.
Visuals that are not self-explanatory make me feel dumb.
gavmor 25 days ago [-]
We don't know what to label those features/dimensions, because they're a reduction form higher dimensions that we also didn't bother to interrogate.
It's possible to figure them out. I wish OP would.
yoouareperfect 25 days ago [-]
OP here, Is there a way to figure that out?
gavmor 25 days ago [-]
(Not OP) I can think of a convoluted and expensive pair-wise comparison method, but I hope there's also a way to figure this out during the application of principal component analysis in a way I don't understand.
Edit: I'm thinking it can't be done without experimentation on the embedding model.
Edit2: Ah, even that might not yield results, because as the basis is derived interstitially through computation, there's no guarantee the features of the final coordinate system will have any accessible relationship to those of the initial basis.
tmshapland 25 days ago [-]
Really neat! We were Tule, in the industrials part of the map in grey.
There's something wonky when I zoom in on Chrome on my laptop. It abruptly shifts to another part of the map.
kure256 25 days ago [-]
Love that, what are Axes Y and C?
DrawTR 25 days ago [-]
Apparently inspired by a comment on this very post! (Above yours, right now.)
> Cool project, but missed opportunity to name the arbitrary dimensions Y and C...
Wow. This is amazing. Extremely practical to use, I'm glad I checked H.N. yesterday.
woodylondon 25 days ago [-]
Really nice to see - also, It would be great when filtering if there was a tabular view at the bottom as well.
uncomplexity_ 25 days ago [-]
hella nice mate very interesting
what's the x and y axes?
jerrygenser 25 days ago [-]
they don't have meaning by themselves. they are two dimensions that umap projected the original embeddings down to in order to show a combination of local neighborhood similarity or closenes
gavmor 25 days ago [-]
Well, they do have meaning by themselves, but it's more work to figure that out. All regular, predictable relationships "have" meaning because all meaning is prescribed. And since we've captured many such prescriptions in LLMs, they can do a decent job approximating those.
welder 25 days ago [-]
Company status isn't up to date... I know there's more than 1 public company that went through YC.
yoouareperfect 25 days ago [-]
Check the filters, not all batches are selected as default. Only the latest ones. If you select all of them, then there are many public companies
k-i-r-t-h-i 25 days ago [-]
This is awesome! Are you able to also add F24?
gniting 25 days ago [-]
Nice! What's the tech stack?
yoouareperfect 25 days ago [-]
For scraping and all the processing, typescript.
Embeddings: openai
For visualizing react (nextjs) + plotly (though the lack of mobile zoom makes me question if I should chsnge it)
mring33621 25 days ago [-]
i'd like a filter by target market (US, EU, APAC...)
yoouareperfect 25 days ago [-]
Coming for v2
ksec 25 days ago [-]
I didn't know YC does Government, Healthcare, Industrials, Real Estate and Construction. All these are great sectors and never made the headline.
Oh, and your website has an unchanged Wordpress favicon...
1. How many of these companies made it to series A, series B, etc
2. How many of these companies have > x employees (where x can be 5, 10, 20, etc)
3. How many of these companies had a founder that moved on to something else
This does require a lot more intelligent data scraping or manual data collection though.
Visuals that are not self-explanatory make me feel dumb.
It's possible to figure them out. I wish OP would.
Edit: I'm thinking it can't be done without experimentation on the embedding model.
Edit2: Ah, even that might not yield results, because as the basis is derived interstitially through computation, there's no guarantee the features of the final coordinate system will have any accessible relationship to those of the initial basis.
There's something wonky when I zoom in on Chrome on my laptop. It abruptly shifts to another part of the map.
> Cool project, but missed opportunity to name the arbitrary dimensions Y and C...
https://imgur.com/a/ycombinator-startups-map-iNX8k6M
what's the x and y axes?
For visualizing react (nextjs) + plotly (though the lack of mobile zoom makes me question if I should chsnge it)