TJ Murphy

@teej_m

Head of Data for the largest streaming platform you’ve never heard of. Previously , , .

San Francisco, CA
Vrijeme pridruživanja: studeni 2008.

Tweetovi

Blokirali ste korisnika/cu @teej_m

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @teej_m

  1. Prikvačeni tweet
    21. kol 2019.

    I used to think endianess, the way computers put bits of bytes together, was some nonsense word. It turns out it’s based on Gulliver’s Travels. Two factions feud over which end to break the hard-boiled egg, the big-endians vs the small-endians.

    Prikaži ovu nit
    Poništi
  2. prije 4 sata

    I am going to feel stupid and that’s ok.

    Prikaži ovu nit
    Poništi
  3. prije 4 sata

    I had planned to go deep on ML this year. But a lot of the bleeding edge feels like twisting knobs or memorizing APIs. So while I wait for fancy ML to get easier, I’m switching gears to stats, starting with these. The last stats I did was AP Statistics in 10th grade 😫😫

    Prikaži ovu nit
    Poništi
  4. 31. sij

    The code, for those who want to see my sloppily assembled garbage. This one's for you

    Prikaži ovu nit
    Poništi
  5. 31. sij

    I have strict multi-step working as well. A couple things worth exploring in more depth - * Strict vs loose funnel support * Allowing event metadata filtering * Cycle counting (A->B->A->B->C)

    Prikaži ovu nit
    Poništi
  6. 31. sij

    So far it's scaled up in linear time. All in memory on my local MBP: * 40-50 bytes of RAM per event stored * Query runs in 400-500ns per scanned event Very naive Python implementation, there's plenty of easy optimizations that could be made.

    Prikaži ovu nit
    Poništi
  7. 31. sij

    Built a thing. I used sorted lists of timestamps, partitioned by user & event. For each user, just walk the two lists for events A & B. Returns # start, # finish, avg arrival time For ~100M events: 10% dropoff runs in < 30s 90% runs in < 2s thanks to

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    30. sij

    Dearest friends on growth teams, Don’t do this shit. Hacking the metric instead of the underlying user problems is the fastest path to brand destruction.

    Prikaži ovu nit
    Poništi
  9. 30. sij

    Yeah this didn't age well. was the last single stock I picked, I'm all-in on index funds now. So there's that.

    Poništi
  10. 29. sij

    This is mostly theory crafting but I'm down to build a prototype if I find a good fit for this.

    Prikaži ovu nit
    Poništi
  11. 29. sij

    Is there a good set of data structures and algorithms for doing user funnel analysis? I want to efficiently calculate how many users did A, then did B within N minutes. DBs do okay but range scans can be killer. Bitmap indexes? R-Trees? A bunch of hash tables glued together?

    Prikaži ovu nit
    Poništi
  12. 28. sij

    Pro tip: toil is a great way to persuade management you need better tools. "These 3 folks spend 20 hours a week futzing with data in Excel. We would reduce that to <2 hours if we allocate $15k to use Looker/Mode/Tableau/PowerBI." Much easier sell when you quantify the impact.

    Prikaži ovu nit
    Poništi
  13. 28. sij

    I think about toil for data teams a lot (toil as defined ). Lots of "data analyst" roles are just toil - repetitive, tactical, no judgement report prep. As a leader you can easily 5x their output by managing their toil - delegate, delete, defer.

    Prikaži ovu nit
    Poništi
  14. 28. sij

    It turns out my audience’s #1 interest is dogs, so this is now a dog account. Here is the dog who works next door to me, taking a break after a ruff day.

    Poništi
  15. 27. sij

    Back in Nov, I had an incredible candidate decline a job offer. She was so good, I revamped our hiring to make sure I found more like her. She just got hit by the Quora layoffs. She's interested in a Data Analyst role at a consumer tech company. DM me for an intro.

    Poništi
  16. 27. sij

    For some, spending 500 hours on leetcode can have a lifetime career value of over $1M. If you’re in that bucket, grinding interview problems is one of the most valuable uses of your time.

    Poništi
  17. proslijedio/la je Tweet
    25. sij

    thinkin bout data

    Poništi
  18. 25. sij

    Got a holiday card from and . I’m both delighted (it’s a lovely card) and pissed that I didn’t think about this when I did growth at .

    Poništi
  19. proslijedio/la je Tweet
    24. sij

    "It's easier to hold your principles 100 percent of the time than it is to hold them 98 percent of the time." -- Clayton Christensen, giant of enterprise, RIP

    Poništi
  20. 24. sij

    Just a reminder— I asked Facebook if I could buy user data, they told me no. I had multiple VPN companies, masquerading as analytics companies, beg to sell me user data. I said no. There's someone out there selling your data but it's not who you think.

    Poništi
  21. 24. sij

    I’ve done hiring but there’s nothing more satisfying than putting together a team. I get chills thinking about the sick stuff we will do together that we couldn’t do apart. It’s all coming together.

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·