Redfin Estimate Deceptive to Home Buyers

Redfin launched its own home price “valuation model” in 2015. How does it compare to the Zestimate?

For listings, the Redfin Estimate is just an adjustment of the listing price – it takes what the agent sets as the list price of a home and pads the price to make it slightly different from the list price. If the list price is high, the Redfin estimate will agree with that high price. If it is low, the Redfin estimate will agree again. The Redfin estimate always agrees with the listing price set by the agent!

Here is an example:

Redfin Estimate:

Zillow Zestimate:

Initial List Price: $12,790,000
Redfin estimate: $12,273,429
Zestimate: $1,009,825

The initial list price of $12,790,000 was a typo by the broker. Within few hours it was corrected to $1,279,000. Redfin was susceptible to the fat finger – since it just takes the list price and adds padding. The Zestimate wasn’t impacted, since it is a machine learnt valuation model and an independent assessment of the property value.

After typo was fixed:

Actual List Price: $1,279,000
Redfin estimate: $12,273,429
Zestimate: $1,009,825

Redfin claims they have the best accuracy. This is what is written on the Redfin website: “The Redfin Estimate has the lowest published error rate of any valuation estimate in the U.S., with only a 1.97 percent median error rate for homes that are for sale.”

This is analogous to a student taking an exam with the answers available and taking credit for a 100% passing grade.

Tweets on Yahoo! Homepage

On May 20th, my team launched Tweets on Yahoo! homepage. Check out the media coverage:

@marshallk32m: Twitter’s new partnership with Yahoo looks like a cool model for a whole lot of media companies … via @prsarahevans

@agillmer_musing8m: Smart move by Yahoo & Twitter to introduce personalized tweets to the masses.

Yahoo! Homepage Redesign

Yahoo! Homepage redesign launched on Feb 20, 2013. We’ve launched a fresh, dynamic and personalized Yahoo! experience that brings you relevant content even faster. It’s now easier for you to get your daily dose of weather, stock quotes, sports scores, and more. These changes are just the first step in making Yahoo! personalized, and the more you visit Yahoo!, the better it will become.

How to Measure Anything

Read a book recently titled “How to Measure Anything” by Douglas W. Hubbard. Here are some notes from the book:

Think of a measurement as “error reduction”

Definition of Measurement – A set of observations that reduce uncertainty where the result is expressed as a quantity.

Thus – a measurement doesn’t have to eliminate uncertainty after all; a mere reduction can be much more than the cost of the measurement. Uncertainty needs to be quantified, but not the observation.

Object of measurement – need to clarify what you want to measure; must be clearly defined.

Rule of Five – 93% chance that the median of a population is between the smallest and largest values in any random sample of five from that population. There are a lot of problems where the Rule of Five really does reduce uncertainty.

Clarifying the Measurement Problem
1) What is the decision this is supposed to support?
2) What really is the thing being measured?
3) Why does this thing matter to the decision being asked?
4) What do you know about it now?
5) What is the value to measuring it further?

What do you mean by it? What does improved it look like?

90% Confidence Interval – Provide upper and lower bound such that you are 90% sure the expected value falls within this range.

Calibration tests improve ability to be a good calibrator.

Measuring Value of Information
If we could measure the value of information itself, we could use that to determine the value of conducting measurements.

We make better decisions when we can reduce uncertainty (make measurements) about them.

Future of Social in Mobile

As more people access social networks from smartphones and tablets, they’ll realize these devices do more than make sharing easier and content creation better. They also allow for instantaneous connections. This real-time self-expression was essentially non-existent before the rise of mobile devices — and now it’s the norm. We’ve become accustomed to knowing what’s going on with our friends and family or favorite celebrities and organizations at the moment it’s happening. This allows us to feel as though we’re experiencing these activities with them, radically changing how we feel and, ultimately, connect.

Technology Behind Yahoo! Social

Yahoo! Mixer is a fully asynchronous service that requires significantly fewer threads than thread-per-connection or thread-per-request architectures like Standard Apache Tomcat applications which don’t scale well with large number of connections/requests. Because of this, the Yahoo! Mixer can efficiently handle a large number of outstanding requests to slower services like Facebook with just a small number of threads tuned to match the number of cores on the server.

Read More on YDN Blog:

Memcached: Slabs, Pages, Chunks

Memcached, a distributed cache for data / object storage in memory, is very important for scaling and improving response time performance. This is especially important for Social websites, which heavily depend on slow outbound data services (e.g. Facebook, database, Identity services, etc.).

To mitigate overhead of heap memory allocation / deallocation, memcached organizes itself into slabs, pages, and chunks. Here I draw a fictitious example in illustration:

At startup, the specified maximum amount of RAM isallocated for the cache with a growth factor.

For instance, the command “memcached -m 64 -f 3” allocates 64 MB in RAM with growth factor of 3. The default smallest chunk size is 96 bytes. If each page is 1 MB, then approx 64 pages will be initially free and not assigned to any slab.

When code writes data into cache, 1 of the free pages will be allocated to slab class. Saw we write 50 byte data into cache, this will be stored in a 96 byte chunk inside. All cache data – key + value + overhead – is stored inside a chunk. 1 free page will be allocated to a slab class that only holds 96 byte chunks.

Here is a short blog by Mike Perham on concepts of slabs, pages and chunks in memcached distributed in-memory cache store:

This blog explains LRU in memcached:

Database Sharding

Nice article on database shard partitioning: