Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Making Scores with HiScore
Search
Hakka Labs
February 13, 2015
Programming
0
3.4k
Making Scores with HiScore
Video here:
Hakka Labs
February 13, 2015
Tweet
Share
More Decks by Hakka Labs
See All by Hakka Labs
New Workflows for Building Data Pipelines
hakka_labs
0
2.9k
Collaborative Topic Models for Users and Texts
hakka_labs
0
2.8k
Groupcache with Evan Owen
hakka_labs
2
5.4k
Testing Android at Spotify
hakka_labs
1
4.5k
It's Not a Bug, It's a Feature!
hakka_labs
0
3.2k
K-means Clustering to Understand Your Users
hakka_labs
0
2k
Building Amy: The Email-based Virtual Assistant by x.ai
hakka_labs
0
5k
Deep Learning and NLP Applications
hakka_labs
3
13k
Go and the Gophers
hakka_labs
2
11k
Other Decks in Programming
See All in Programming
フロントエンド開発のためのブラウザ組み込みAI入門
masashi
5
2.3k
いま中途半端なSwift 6対応をするより、Default ActorやApproachable Concurrencyを有効にしてからでいいんじゃない?
yimajo
2
440
NixOS + Kubernetesで構築する自宅サーバーのすべて
ichi_h3
0
1k
CSC509 Lecture 05
javiergs
PRO
0
300
スキーマ駆動で、Zod OpenAPI Honoによる、API開発するために、Hono Takibiというライブラリを作っている
nakita628
0
180
Domain-centric? Why Hexagonal, Onion, and Clean Architecture Are Answers to the Wrong Question
olivergierke
3
910
Six and a half ridiculous things to do with Quarkus
hollycummins
0
180
なぜGoのジェネリクスはこの形なのか? Featherweight Goが明かす設計の核心
ryotaros
7
1.1k
Writing Better Go: Lessons from 10 Code Reviews
konradreiche
0
1.6k
Things You Thought You Didn’t Need To Care About That Have a Big Impact On Your Job
hollycummins
0
230
CSC509 Lecture 06
javiergs
PRO
0
260
CSC305 Lecture 04
javiergs
PRO
0
270
Featured
See All Featured
Producing Creativity
orderedlist
PRO
347
40k
The Cost Of JavaScript in 2023
addyosmani
55
9k
Agile that works and the tools we love
rasmusluckow
331
21k
4 Signs Your Business is Dying
shpigford
185
22k
Code Review Best Practice
trishagee
72
19k
What’s in a name? Adding method to the madness
productmarketing
PRO
24
3.7k
Embracing the Ebb and Flow
colly
88
4.9k
It's Worth the Effort
3n
187
28k
The Power of CSS Pseudo Elements
geoffreycrofte
79
6k
What's in a price? How to price your products and services
michaelherold
246
12k
Designing for Performance
lara
610
69k
How To Stay Up To Date on Web Technology
chriscoyier
791
250k
Transcript
Making Scores with HiScore Abe Othman
None
None
None
None
HiScore is a python library for creating and maintaining scores
It uses a novel quasi-Kriging solution to a new methodology,
supervised scoring
What are scores?
Scores are a tool for domain experts to communicate their
expertise to a broad audience
88 51 27
} 58 Score Function Dimensions Score
There is no one correct scoring function
Scores are typically developed using the dual approach
1. Select a set of basis functions f(x ⃗) =
∑ γᵢφᵢ(x ⃗)
2. Adjust coefficients until things look right f(x ⃗) =
∑ γᵢφᵢ(x ⃗)
Dual scores ossify
Walkscore Problems Score of 100, but the highest crime in
SF
Supervised scoring: a primal approach
Experts start by labeling a reference set and the objects’
dimensions
Algorithm makes a scoring function that interpolates and obeys the
monotone relationship
Some nice features
Monotonicity is important for score acceptance and understanding
See a mis-scored point? Add it to the reference set
and re-run!
OK, but what algorithm?
Easy in one dimension
None
None
None
Hard in many dimensions
Failed approach: simplical interpolation
None
Failed approach: B-spline product bases
Supervised Scoring with Monotone Multidimensional Splines, AAAI 2014
Curse of dimensionality!
None
None
None
Failed approach: RBF with monotone row generation constraints
Failed approach: Neural Networks
None
None
Success: Beliakov
Reminder: Lipschitz Continuity |f(a)-f(b)| < C |a-b|
None
Monotone Lipschitz continuity
None
1. Project monotone Lipschitz cones from each point to generate
upper and lower bounds
2. Find the sup and inf constraints from the bounding
cones
3. Function value is halfway in-between the sup and inf
bounds
Beliakov example
Beliakov plateaux
Beliakov plateaux
How can we smooth and improve this?
Abandon Lipschitz, just project minimal cones from each point
None
`
HiScore solution
Using HiScore: Simplified Water Well Score
None
None
Two factors: Distance from nearest latrine and platform size
Label a reference set by taking high, middle and low
values in each dimension
Distance: 0m, 10m, 50m Size: 1SF, 25SF, 100SF
Score Distance Size 0 0 1 5 0 25 10
0 100 20 10 1 50 10 25 60 10 100 65 50 1 90 50 25 100 50 100 Monotone Relationship: (+, +)
import hiscore reference_set = {(0,1): 0, (0,25): 5, (0,100): 10,
(10,1): 20, (10,25): 50, … } mono_rel = [1,1] hiscore.create(reference_set, mono_rel, minval=0, maxval=100)
None
Complicate the model with additional factors
Avoid curse of dimensionality by building a tree
None
Possible to easily construct and understand scores with dozens of
input dimensions
Making dimensions monotone: blood pressure
None
S+ > 0 S- = 0 D+ > 0 D-
= 0 D+ = 0 D- > 0 S+ = 0 S- > 0
What do you want to score? github.com/aothman/ hiscore $ pip
install hiscore
Thanks!
[email protected]