The Unofficial Google Data Science Blog

JUNE 30, 2016

A naïve comparison of the exposed and unexposed groups would produce an overly optimistic measurement of the effect of the ad, since the exposed group has a higher baseline likelihood of purchasing a pickup truck. It should be noted that inverse probability weighting is not generally optimal (i.e., the curse of dimensionality).

Statistics 44

Statistics Optimization Modeling Experimentation 44

The Unofficial Google Data Science Blog

OCTOBER 7, 2015

by HENNING HOHNHOLD, DEIRDRE O'BRIEN, and DIANE TANG In this post we discuss the challenges in measuring and modeling the long-term effect of ads on user behavior. Nevertheless, A/B testing has challenges and blind spots, such as: the difficulty of identifying suitable metrics that give "works well" a measurable meaning.

Modeling 61

Modeling Experimentation Knowledge Discovery Testing 61

Occam's Razor

MARCH 12, 2012

All the way back in 2007, I was evangelizing the value of moving away from the "small data" world of clickstream data to the "bigger data" world of using multiple data sources to make smarter decisions on the web. Here's the "bigger web analytics data" picture from 2007… Multiplicity!

Big Data 127

Big Data Data-driven Unstructured Data Marketing 127

IBM Big Data Hub

JUNE 1, 2023

The rule proposal would require US publicly traded companies to disclose annually how their businesses are assessing, measuring and managing climate-related risks. This would include disclosure of greenhouse gas emissions as a measure of exposure to climate-related risk.

Risk 58

Risk Reporting Measurement Consulting 58

Edwin Chen

MARCH 13, 2011

This is a summary of Bell and Koren’s 2007 Scalable Collaborative Filtering with Jointly Derived Neighborhood Interpolation Weights paper. We’ll see how using an optimization method to derive weights (as opposed to deriving weights via a similarity function) overcomes these two limitations.

Optimization 40

Optimization Modeling Cost-Benefit Measurement 40

Occam's Razor

OCTOBER 15, 2012

I spend 70% of my time in the US and for those discussions I'm primary looking at speed (connection above), mobile penetration (yes, 2007 was the year of mobile!), It is SUPER important to passionately measure the offline impact of your online efforts until the day you die (cue, dramatic music). It is an intense amount of effort.

Advertising 136

Advertising Marketing Strategy Insurance 136

The Unofficial Google Data Science Blog

APRIL 19, 2021

The numerical value of the signal became decoupled from the event it was measuring even as the ordinal value remained unchanged. isn’t good enough: it optimizes the calibration term, but pays the price in sharpness. In practice, we enforce this by optimizing over $log(beta_i)$. And users may start receiving a lot more spam!

Modeling 122

Modeling IT Metrics Testing 122

The Unofficial Google Data Science Blog

JULY 22, 2020

For this reason we don’t report uncertainty measures or statistical significance in the results of the simulation. Ramp-up solution: measure epoch and condition on its effect If one wants to do full traffic ramp-up and use data from all epochs, they must use an adjusted estimator to get an unbiased estimate of the average reward in each arm.

Experimentation 95

Experimentation Statistics Testing Strategy 95

Occam's Razor

FEBRUARY 7, 2011

Brian Krick: Best way to measure and communicate "available demand" from available channels (social, search, display) for forecast modeling. Additionally, it is exceptionally difficult to measure available demand because 1. Alex Cohen: How to optimize with sparse data! No accountability = very little desire to measure.

Analytics 94

Analytics Measurement Metrics Predictive Analytics 94

Domino Data Lab

FEBRUARY 4, 2019

The ability to measure results (risk-reducing evidence). Frédéric Kaplan, Pierre-Yves Oudeyer (2007). Oudeyer and Twomey separately describe processes for curiosity-driven exploration which have parallels in A* search and heuristic optimizations for gradient descent. You’ll need to read the papers.

Data Science 66

Data Science Experimentation Data-driven Machine Learning 66

Domino Data Lab

SEPTEMBER 25, 2019

We see it when working with log data, financial data, transactional data, and when measuring anything in a real engineering system. Fortunately, the forecast package has a number of functions to make working with time series data easier, including determining the optimal number of diffs. 2007-01-04 34.50 2007-01-05 33.96

Forecasting 87

Forecasting Modeling Statistics Optimization 87