Slide 1

Slide 1 text

Adding machine learning to a web app Richard Dallaway @d6y

Slide 2

Slide 2 text

It’s easy, right? 1. Get some data 2. Find magic stats & algorithms 3. Predict the future!

Slide 3

Slide 3 text

It’s easy, right? 1. Get some data 2. Find magic stats & algorithms 3. Predict the future! that helps solve the problem frame the problem explore the data pilot what’s success? does this help users? UI impact? behaviour changes?

Slide 4

Slide 4 text

Make it easier for users to #tag todos

Slide 5

Slide 5 text

No content

Slide 6

Slide 6 text

tag

Slide 7

Slide 7 text

Can we suggest tags as a task is typed in?

Slide 8

Slide 8 text

Can we suggest tags as a task is typed in? useful

Slide 9

Slide 9 text

Where to start? from “Algorithms of the Intelligent Web”

Slide 10

Slide 10 text

“Google uses Bayesian filtering the way Microsoft uses the if statement" as told to Joel Spolsky

Slide 11

Slide 11 text

A contrived example #home 40% #work 60% “windows” 50% “fix” 50% “windows” 16.6% “fix” 83.4% p(#home | “fix”, “windows”) ?

Slide 12

Slide 12 text

A contrived example #home 40% #work 60% “windows” 50% “fix” 50% “windows” 16.6% “fix” 83.4% p(#home | “fix”) = 40% x 50% (40% x 50%) + (83.4% x 60%) = about 28%

Slide 13

Slide 13 text

A contrived example #home 28% #work 72% “windows” 50% “fix” 50% “windows” 16.6% “fix” 83.4% p(#home | “fix”, “windows”) = 28% x 50% (28% x 50%) + (16.6% x 72%) = about 55% p(#home | “fix”)

Slide 14

Slide 14 text

p(C | e) = P(C) x P(e | C) P(e)

Slide 15

Slide 15 text

p(C | e) = P(C) x P(e | C) P(e) “the estimation of P(e | C) can be viewed as the central issue in designing learning systems. ” — Weiss & Kulikowski “Computer Systems that Learn”

Slide 16

Slide 16 text

Demo: addsharedo.com with tag suggestions

Slide 17

Slide 17 text

results of running the model

Slide 18

Slide 18 text

It’s easy, right? 1. Explore the data 2. Frame your problem 3. Measure the performance honestly

Slide 19

Slide 19 text

Write it yourself? Find a library? Ask Google to do it for you? How to build? ...but pilot with offline data first.

Slide 20

Slide 20 text

www.manning.com 40% off with bathcamp40 until August 10th

Slide 21

Slide 21 text

Questions? or later: @d6y [email protected]