## Hitting the Books: EADS Summer School on Hashing

Rob, Matt, and I just wrapped up our trip to Copenhagen for the EADS Summer School on Hashing at the University of Copenhagen and it was a blast! The lineup of speakers was, simply put, unbeatable: Rasmus Pagh, Graham Cormode, Michael Mitzenmacher, Mikkel Thorup, Alex Andoni, Haim Kaplan, John Langford, and Suresh Venkatasubramanian. There’s a good chance that any paper you’ve read on hashing, sketching, or streaming has one of them as a co-author or is heavily influenced by their work. The format was three hour-and-a-half lectures for four days, with exercises presented at the end of each lecture. (Slides can be found here. They might also post videos. UPDATE: They’ve posted videos!)

Despite the depth of the material, almost all of it was approachable with some experience in undergraduate math and computer science. We would strongly recommend both of Michael Mitzenmacher’s talks (1, 2) for an excellent overview of Bloom Filters and Cuckoo hashing that are, in my opinion, significantly better and more in depth than any other out there. Specifically, the Bloom Filter talk presents very elegantly the continuum of Bloom Filter to Counting Bloom Filter to Count-Min Sketch (with “conservative update”) to the Stragglers Problem and Invertible Bloom Filters to, finally, extremely recent work called Odd Sketches.

Similarly, Mikkel Thorup’s two talks on hashing (1, 2) do a very thorough job of examining the hows and whys of integer hashing, all the way from the simplest multiply-mod-prime schemes all the way to modern work on tabulation hashing. And if you haven’t heard of tabulation hashing, and specifically twisted tabulation hashing, get on that because (1) it’s amazing that it doesn’t produce trash given how simple it is, (2) it’s unbelievably fast, and (3) it has been proven to provide the guarantees required for almost all of the interesting topics we’ve discussed on the blog in the past: Bloom Filters, Count-Min sketch, HyperLogLog, chaining/linear-probing/cuckoo hash tables, and so on. We really appreciated how much attention Mikkel devoted to practicality of implementation and to empirical performance when discussing hashing algorithms. It warms our heart to hear a leading academic in this field tout the number of nanoseconds it takes to hash an item as vocally as the elegance of the proof behind it!

We love this “Summer School” format because it delivers the accumulated didactic insight of the field’s top researchers and educators to both old techniques and brand new ones. (And we hope by now that everyone reading our blog appreciates how obsessed we are with teaching and clarifying interesting algorithms and data structures!) Usually most of this insight (into origins, creative process, stumbling blocks, intermediate results, inspiration, etc.) only comes out in conversation or lectures, and even worse is withheld or elided at publishing time for the sake of “clarity” or “elegance”, which is a baffling rationale given how useful these “notes in the margin” have been to us. The longer format of the lectures really allowed for useful “digressions” into the history or inspiration for topics or proofs, which is a huge contrast to the 10-minute presentations given at a conference like SODA. (Yes, obviously the objective of SODA is to show a much greater breadth of work, but it really makes it hard to explain or digest the context of new work.)

In much the same way, the length of the program really gave us the opportunity to have great conversations with the speakers and attendees between sessions and over dinner. We can’t emphasize this enough: if your ambition to is implement and understand cutting edge algorithms and data structures then the best bang for your buck is to get out there and meet the researchers in person. We’re incredibly lucky to call most of the speakers our friends and to regularly trade notes and get pointers to new research. They have helped us time and again when we’ve been baffled by inconsistent terminology or had a hunch that two pieces of research were “basically saying the same thing”. Unsurprisingly, they are also the best group of people to talk to when it comes to understanding how to foster a culture of successful research. For instance, Mikkel has a great article on how to systematically encourage and reward research article that appears in the March 2013 issue of CACM (pay-wall’d). Also worthwhile is his guest post on Bertrand Meyer’s blog.

If Mikkel decides to host another one of these, we cannot recommend attending enough. (Did we mention it was free?!) Thanks again Mikkel, Rasmus, Graham, Alex, Michael, Haim, and John for organizing such a great program and lecturing so eloquently!

## Writing Analytics SQL with Common Table Expressions

Author’s Note: Hello readers! I’m Josh O’Brien. I recently joined the Science team as a junior engineer, and this is my first post for the blog.

### Introduction

One of my first tasks with the Science team has been learning to write effective analytics SQL. I came in with a basic knowledge of SQL, but writing complex analytics reports required me to learn tools and strategies for managing complexity that aren’t yet part of the standard introductions to SQL. Luckily, I had the Science team to teach me to work with Common Table Expressions (CTEs). I’ve come to love CTEs for the clarity that they’ve helped bring to my thinking and writing in SQL. The CTE syntax encourages me to reason through a problem as a sequence of simple parts and enables me to directly code a solution in terms of those parts, which I can individually document and test for correctness. Working with CTEs has jump-started my productivity, and helped the team as a whole set a higher standard for our SQL.

In the Science team’s experience, much of the common frustration with SQL comes down to a failure to treat SQL queries as declarative programs that demand the same care as imperative programs. SQL is code, and we should treat it as such. We can better manage the complexity of SQL by using the same basic techniques we do in other languages: we can divide work into composable parts, document our intent, and test for correctness. We use CTEs as a foundation for building queries that are factored, documented, and tested, and we’ve enjoyed excellent results writing and maintaining numerous hundred- and thousand-line reports using this approach.

In this post, I’ll share an example of how the Science team uses CTEs to treat SQL as code. I’ll walk through the process of writing an analytics report with CTEs, and show how CTEs help me think through a problem and implement, document, and test a solution.

* If you’re thinking that CTEs are no better than temporary tables or views for these purposes, read on. CTEs, temporary tables, and views all have their place in our SQL toolkit. We use CTEs because they are best suited for this work. For more on the relative merits of CTEs, temporary tables, and views, please see the appendices to this post.

### Common Table Expressions

Before we dive into the example report, let’s take a quick look at the CTE syntax we’ll be using. CTEs are defined inside of a WITH clause attached to a primary statement. Within the scope of the larger query, each CTE can be manipulated like a table. This allows us to chain CTEs together and build sequences of operations. In the following diagram, we’re building up a four-part query, part by part. We start with two parts: a foo CTE attached to a main SELECT statement. Next, we add a bar CTE. In the final step, we add a baz CTE to complete the four-part query.

Examples of two-, three-, and four-part queries with Common Table Expressions. The query grows by one CTE at each stage.

Notice what we did here. In the foo, bar, and baz CTEs, we now have three intermediate result sets that we can test individually and “print” with a SELECT *. Once we know each part is correct, we add another, until we’ve solved our problem. We can use CTEs to break queries into as many simple parts as the problem requires.

We use CTEs rather than temporary tables or views to decompose queries in development because they are simpler to use. There is no need to add the complexity of managing CREATE and DROP statements at this stage in the writing process.

### Frequency Report

We’ll use a simplified example report to illustrate how we use CTEs in our everyday work: a frequency report. A frequency report is an online advertising analytics report that helps advertisers determine the number of ads to serve users over a specific time period. Advertisers want to reach out to customers enough times to build awareness of and interest in their offerings, but not so many times that customers become jaded or annoyed. A frequency report breaks down return on advertising investment by the number of ads users have been shown, a classification known as a user’s impression frequency class.

This report produces data that can be graphed as:

An example of a report in our UI, showing impressions for an advertiser by frequency class.

Stripped all the way down, the basic query that generates the report above is:

WITH impression_counts AS (
SELECT user_id,
SUM(1) AS impression_count
FROM impressions
GROUP BY 1
)
SELECT impression_count      AS frequency_class,
SUM(impression_count) AS total_impressions
FROM impression_counts
GROUP BY 1
;


The challenge of writing these reports comes from managing all the additional data we need. Actual reporting queries need to correctly handle the complexity of timestamp, ad campaign, conversion attribution, click, and cost data without becoming tangled messes.

For this simplified example, we’ll start with tables recording impression (ad view), click (ad interaction), and conversion (sale) events, and produce a frequency report tracking the total number of users, impressions, clicks, and conversions for each impression frequency class for each ad campaign in the database for the month of March 2014. We can visualize our task like this:

Our task: use the impressions, clicks, impression_attributed_conversions, and click_attributed_conversions tables to produce a frequency report for the month of March 2014.

### Thinking with CTEs

Working with CTEs begins with reasoning about the problem in terms of the stages and parts needed to produce the report. From the above starting point, we can already work out four main stages.

We’ll need to:

• FILTER the four input tables by record_date,
• GROUP BY user_id and campaign_id, and SUM to get user-level counts for impressions, clicks, and conversions,
• JOIN those counts together on user_id and campaign_id, and finally,
• GROUP BY impression_count (= frequency_class) and campaign_id, and SUM to generate the report totals for users, impressions, clicks, and conversions.

We can express the relationships between these operations visually:

A map of the query to produce the frequency report. Each of the conceptual parts connecting the green input tables to the orange frequency report will be written as a simple CTE.

In one form or another, each of these operations would need to be a part of any query that produces this report. With CTEs, we can preserve the logical clarity of our thought process in the code itself. Each of the main parts of this query will be implemented using simple CTEs that serve only one main purpose. For added clarity, we will name and comment the CTEs to communicate our intent at every stage. This technique yields a query that we can read straight through and maintain with ease, just like our other code.

### Writing with CTEs

Let’s take a look at a CTE from each stage right now. The full query with documentation comments can be found here, and in the appendices to this post.

First come the three filter CTEs. Here’s the CTE for filtered_impressions. Its only purpose is to filter the impressions table down to March 2014:

filtered_impressions AS (
SELECT record_date,
user_id,
campaign_id
FROM impressions
WHERE record_date >= '2014-03-01' AND
record_date <  '2014-04-01'
)


Next, we calculate user-level counts for impressions, clicks, and conversions. Each of the three “counts” CTEs performs only a simple aggregate function: a GROUP BY and a SUM. Here is the impression_counts CTE:

impression_counts AS (
SELECT user_id,
campaign_id,
SUM(1) AS impression_count
FROM filtered_impressions
GROUP BY 1, 2
)


After that, we JOIN the three “counts” CTEs together in a single long table. This collated_counts CTE is the longest in the query, but, like the others, it has only one main purpose:

collated_counts AS (
SELECT imp.user_id           AS user_id,
imp.campaign_id       AS campaign_id,
imp.impression_count  AS impression_count,
cl.click_count        AS click_count,
conv.conversion_count AS conversion_count
FROM impression_counts imp
LEFT OUTER JOIN click_counts cl ON
imp.user_id      = cl.user_id AND
imp.campaign_id  = cl.campaign_id
LEFT OUTER JOIN conversion_counts conv ON
imp.user_id      = conv.user_id AND
imp.campaign_id  = conv.campaign_id
)


Last comes the main SELECT statement. Its only purpose is to group by impression_count (= frequency_class) and campaign_id, and calculate the four SUMs for the report:

SELECT impression_count                   AS frequency_class,
campaign_id                        AS campaign_id,
SUM(1)                             AS total_users,
SUM(impression_count)              AS total_impressions,
SUM(COALESCE(click_count, 0))      AS total_clicks,
SUM(COALESCE(conversion_count, 0)) AS total_conversions
FROM collated_counts
GROUP BY 1, 2


### Testing with CTEs

As we build up the query with CTEs, we leverage the ability to SELECT from each CTE individually to test for correctness as part of the writing process. This basic testing can be as simple as three files in a text editor, which we execute from psql (or equivalent) in a sequence as we write:

• setup.sql: CREATE tables and INSERT rows of test data
• test.sql: the query itself
• teardown.sql: DROP the tables created in setup.sql

We write and comment one CTE at a time in the test file. Each time we add a CTE, we add test rows to exercise that CTE to the setup file, and include comments to indicate what should happen to those rows when we SELECT * from the relevant CTE. When the output matches our expectations, we move to the next part of the query, and repeat the process.

As an example, initial tests for the filtered_impressions CTE could consist of creating an impressions table and inserting five rows to exercise the date range in the WHERE clause. We indicate our expectations for those rows with brief comments:

CREATE TABLE impressions (
record_date  date   NOT NULL,
user_id      bigint NOT NULL,
campaign_id  bigint NOT NULL
);
INSERT INTO impressions (record_date, user_id, campaign_id) VALUES
/* The following 2 rows should not appear in filtered_impressions: */
('2014-02-28', 707, 7),
('2014-04-01', 707, 7),
/* The following 3 rows should appear in filtered_impressions: */
('2014-03-01', 101, 1),
('2014-03-15', 101, 1),
('2014-03-31', 101, 1)
;


This basic testing at the time of writing is not a substitute for a comprehensive test framework, but it is enough to catch many errors that could otherwise sneak through, and it provides a good return on a modest investment of effort. By the time the full query is complete, this process will have generated tests and documentation for each part of the query.

### Conclusion

This method of working with CTEs has helped me by bringing clarity and simplicity to complex analytics queries. Thinking, writing, and testing with CTEs helps me treat SQL as part of software engineering practice by writing SQL that’s factored, documented, and tested more like other code.

The Science team thinks of this method as producing a foundation for further refinements. When appropriate, optimizations for performance can and will be made, but we focus on correctness first. Optimizations tend to add complexity, and before we do that, we want to mitigate the complexity of the query as much as possible.

By starting with CTEs, we can more easily write queries that we can quickly read and reuse six months from now. Analysts can return to their models and analyses with confidence and engineers are better able to add new features to reports without introducing new bugs. We’re building upon a foundation of factored, documented, and tested SQL.

### Appendices

##### On CTEs, Temporary Tables, and Views

We asked Christophe Pettus of PostgreSQL Experts to help illuminate the tradeoffs between CTEs, views, and temporary tables, and received the following helpful response, which we publish here with his permission and our thanks:

[E]ach have characteristics that can make them better or worse in particular situations:

1. CTEs are optimization fences; the query planner will plan CTEs
separately from the rest of the query. This can be good or bad,
depending on the way the CTE is used.

2. Views are *not* optimization fences; you can think of them as being
textually inserted into the query at the appropriate place, so queries
can be rewritten, join clauses moved around, etc.

3. Temporary tables can have indexes; for very large intermediate result
sets, this can be essential for good performance.

We agree that the choice between CTEs, temporary tables, and views is a matter of balancing the different trade-offs of the different stages of software development.

As explained in this post, the Science team finds the balance in favor of CTEs as the foundation for query development. We reach for the CTE syntax first for its clarity and ease of use. When we write and test queries part-by-part, we want to keep the code as clear and simple as possible. Juggling extra CREATE and DROP statements for temporary tables or views works against that goal.

Once we have a correct, clear foundation, then we move onto the optimizations I mentioned in the conclusion. At that point, we consider re-writing CTEs as views or materialized tables on a case-by-case basis. Sometimes the balance tips away from CTEs. In our experience, the most common reason for this has been to gain the performance benefits of indexing on intermediate result sets that can contain hundreds of millions to tens of billions of rows.

## Open Source Release: java-hll

We’re happy to announce our newest open-source project, java-hll, a HyperLogLog implementation in Java that is storage-compatible with the previously released postgresql-hll and js-hll implementations. And as the rule of three dictates, we’ve also extracted the storage specification that makes them interoperable into it’s own repository. Currently, all three implementations support reading storage specification v1.0.0, while only the PostgreSQL and Java implementations fully support writing v1.0.0. We hope to bring the JS implementation up to speed, with respect to serialization, shortly.

## AK at re:Invent 2013

I was given the opportunity to speak at AWS re:Invent 2013, on Wednesday, Nov. 13th, about how we extract maximum performance, across our organization, from Redshift. You can find the slides here, and my voice-over, though not identical, is mostly captured by the notes in those slides.

The talk, in brief, was about the technical features of Redshift that allow us to write, run, and maintain many thousands of lines of SQL that run over hundreds of billions of rows at a time. I stayed away from nitty-gritty details in the talk, since I only had 15 minutes, and tried to focus on high-level take-aways:

• SQL is code, so treat it like the rest of your code. It has to be clean, factored, and documented. Use CTEs to break queries into logical chunks. Use window functions to express complex ideas succinctly.
• Redshift is an MPP system with fast IO, fast sorting, and lots of storage. Use this to your advantage by storing multiple different sort orders of your data if you have different access patterns. Materialize shared intermediates so many queries can take advantage of them.
• Redshift has excellent integration with S3. Use the fat pipes to cheaply materialize query intermediates for debugging. Use the one-click snapshot feature to open up experimentation with schema, data layout, and column compression. If it doesn’t work out, you revert to your old snapshot.
• Use the operational simplicity of Redshift to be frugal. Turn over the responsibility of managing cluster lifecycles to the people that use them. For instance, devs and QA rarely need their clusters when the workday is done. The dashboards are such a no-brainer that it’s barely a burden to have them turn off and start up their own clusters. Allow users to take responsibility for their cluster, and they will become more responsible about using their cluster.
• Use the operational simplicity of Redshift to be more aware. With just a few clicks, you can launch differently sized clusters and evaluate your reports and queries against all of them. Quantify the cost of your queries in time and money.
• It’s a managed service: stop worrying about nodes and rows and partitions and compression! Get back to business value:
• How long does the customer have to wait?
• How much does this report cost?
• How do I make my staff more productive?
• How do I minimize my technical debt?

## Sketch of the Day: Frugal Streaming

We are always worried about the size of our sketches. At AK we like to count stuff, and if you can count stuff with smaller sketches then you can count more stuff! We constantly have conversations about how we might be able to make our sketches, and subsequently our datastores, smaller. During our science summit, Muthu pointed us at some of the new work in Frugal Streaming. The concept of Frugal Streaming is to process your data as it comes, O(N), but severely limit the amount of memory you are using for your sketch, and by “severely” they mean using perhaps one or two pieces of information. Needless to say, we were interested!

### History

The concept of Frugal Streaming reminds me of an algorithm for finding the dominant item in a stream, MJRTY written by Boyer and Moore in 1980. (This paper has a very interesting history). The MJRTY algorithm sets out to solve the problem of finding the majority element in a stream (an element comprising more than 50% of the stream). Moore proposed to solve this by using only 2 pieces of information and a single scan of the data.

Imagine you have a stream of names (“matt”, “timon”, “matt”, “matt”, “rob”, “ben”, …) and you wanted to know if any name appeared in more than half the stream. Boyer and Moore proposed the following:

count = 0
majority = ""

for val in stream:
if count == 0:
majority = val
count = 1
elif val == majority:
count += 1
else:
count -= 1

print majority if count > 0 else "no majority!"


If you’re anything like me you will go through a few phases: “That can’t work!”, “Wait, that works?!”, “Nah, this data will break that”, “Holy sh*t that works!!”. If you think about it for a minute you realize that it HAS to work. If any element in the stream comprises more than half the stream values there is no way to get to the end and have a counter of zero. To convince yourself suppose the majority element only comprises the first half + 1 of your N-length stream. The counter would count up to $N/2+1$ and then start decrementing until all N values have been seen, which would leave the counter at $2 = (N/2+1) - (N/2-1)$*. This will hold regardless of the ordering of the elements in the stream. A simple simulation is provided by the authors. Philippe Flajolet apparently “liked this algorithm very much and called it the ‘gang war’, because in his mind, every element is a gang member, and members of opposite gangs are paired in a standoff, and shoot each other. The gang members remaining are the more numerous”**.

The astute among you will have noticed that this algorithm only works if there is, in fact, a majority element. If there is not then it will fail. A stream such as {“matt”,”matt”,”timon”,”timon”,”rob”} would result in the algorithm returning “rob”. In practice you need ways of ensuring that your stream does indeed have a majority element or have a guarantee ahead of time.

* Note, that formula is for an even length stream. For a stream of odd length the counter will indeed be at 1. Proof is left to the reader.

** Thanks to Jeremie Lumbroso for his insightful feedback on the history of MJRTY and his memory of Philippe’s explanation.

### One “bit” – Frugal-1U

In their Frugal Streaming paper, Qiang and Muthu decided to see if they could find a frugal solution to the streaming quantile problem. The streaming quantiles problem is one I enjoy quite a bit and I have used it as an interview question for potential candidates for some time. Simply stated it is: “How would you find the quantiles of a stream of numbers in O(N) with limited memory?” There are a few different approaches to solve this problem, with the most widely known probably being Count-Min Sketch. However, with Count-Min you need to keep your keys around in order to query the sketch. There are other approaches to this question as well.

Instead of focusing on quantiles generally, Qiang and Muthu’s first solution is a frugal way to find the median of a stream. As with MJRTY above, I’ll just write it down and see how you react:

median_est = 0
for val in stream:
if val > median_est:
median_est += 1
elif val < median_est:
median_est -= 1


Granted, the above is just for the median, where the stream is much longer than the value of the median, but it is so simple that I don’t think I would have ever considered this approach to the problem. The extension to quantiles is equally as shocking. If you are trying to find the 75th percentile of the data stream you do the same as above but increment up randomly 75% of the time and decrement down randomly 25% of the time:


quantile_75 = 0
for val in stream:
r = random()
if val > quantile_75 and r > 1 - 0.75:
quantile_75 += 1
elif val < quantile_75 and r > 0.75:
quantile_75 -= 1


As the paper states:

The trick to generalize median estimation to any $\frac{h}{k}$ -quantile estimation is that not every stream item seen will cause an update. If the current stream item is larger than estimation, an increment update will be triggered only with probability $\frac{h}{k}$. The rationale behind it is that if we are estimating $\frac{h}{k}$ -quantile, and if the current estimate is at stream’s true $\frac{h}{k}$ -quantile, we will expect to see stream items larger than the current estimate with probability $1-\frac{h}{k}$ .

### Finding Quantiles With Two “bits”- Frugal-2U

There are a few obvious drawbacks to the above algorithm. Since we are only incrementing by 1 each time, the time to converge is linear and our initial guess of zero could be very bad. Secondly, and by design, the algorithm has no memory, can fluctuate wildly and, as they show in the paper, the estimates can drift very far away. (Note: while it is extremely unlikely that the estimates will drift far from the correct values the paper has some very loose bounds on how bad it can drift. See Lemma 3 in the paper.) They suggest a few improvements over Frugal-1U where you essentially include a varying step (instead of just incrementing by 1 every time) and 1 “bit” so you know which way you incremented in the last update. The intuition here is that if you have been consistently incrementing up or down for the last few elements of a stream then you are probably “far” away from the quantile in question. If this is the case we can speed up convergence time by incrementing a larger amount. The Frugal-2U algorithm:

def frugal_2u(stream, m = 0, q = 0.5, f = constantly_one):
step, sign = 1, 1

for item in stream:
if item > m and random() > 1 - q:
# Increment the step size if and only if the estimate keeps moving in
# the same direction. Step size is incremented by the result of applying
# the specified step function to the previous step size.
step += f(step) if sign > 0 else -1 * f(step)
# Increment the estimate by step size if step is positive. Otherwise,
# increment the step size by one.
m += step if step > 0 else 1
# Mark that the estimate increased this step
sign = 1
# If the estimate overshot the item in the stream, pull the estimate back
# and re-adjust the step size.
if m > item:
step += (item - m)
m = item
# If the item is less than the stream, follow all of the same steps as
# above, with signs reversed.
elif item < m and random() > q:
step += f(step) if sign < 0 else -1 * f(step)
m -= step if step > 0 else 1
sign = -1
if m < item:
step += (m - item)
m = item
# Damp down the step size to avoid oscillation.
if (m - item) * sign < 0 and step > 1:
step = 1



You can play around with the 1U and 2U algorithms in the simulation below.

Click above to run the Frugal Streaming simulation

As usual, there are a few interesting tidbits as well. If you read the paper you will see that they define the updates to step as a function. This means they are allowing many different types of increments to step. For example, instead of increasing the size of step by 1 each time we could increase it by 10 or even have it increase multiplicatively. They do talk about some uses of different updates to step but there is no analysis around this (yet) and they restrict all of the work in the paper to step increments of 1. We offer a few different step update functions in the simulation and they indeed do interesting things. Further exploration is definitely needed to get some insights here.

A non-obvious thing about the step variable is how it behaves under decrements. My initial thought was that step would get large if your current estimate was far below the actual value (thus allowing you to approach it faster from below) and that step would get to be a large negative number if your current estimate was very far above the actual value. This turns out to just be flat wrong. The negative updates to step have the effect of stabilizing the estimate (notice when step is negative that the updates to your estimates are always ± 1 ). If you read the algorithm closely you will see that step decrements when you consistently alternate up and down updates. This behavior occurs when the estimate is close to the actual value which causes step to become a very large negative number. And I mean VERY large. In practice we have seen this number as small as $-10^{102}$ for some simulations.

### Monitoring

One of the first things I thought of when I saw this sketch was to use it as a monitoring tool. Specifically, perhaps it could be used to replace the monitoring we use on our application server response times. It turns out that using this sketch for monitoring introduces some very interesting issues. So far, we have mostly talked about what I will call “static streams”. These are streams that have data in them which is pulled consistently from some static underlying distribution (as in the examples above). However, what if the underlying distribution changes? For example, what if one of our servers all of the sudden starts responding with slower response times? Does this frugal sketch enable you to quickly figure out that something has broken and fire off an alarm with high confidence? Well, in some ways this is an identical problem to cold start: how long does it take for an initial guess to reach a stable quantile? Unfortunately, there is no easy way to know when you have reached “equilibrium” and this makes identifying when an underlying distribution change has occurred difficult as well. The paper ends with an open challenge:

… as our experiments and insights indicate, frugal streaming algorithms work with so little memory of the past that they are adaptable to changes in the stream characteristics. It will be of great interest to understand this phenomenon better.

The paper shows some interesting experiments with changing data, but I agree with Qiang: we need to understand these algorithms better before I swap out all of our server monitoring tools (and our ops team would kill me). However, since these are so simple to implement and so small, we can easily deploy a few tests and see how the results compare “in the field” (you can play around with this by changing the underlying stream distribution in our simulation above.)

### Conclusion

The frugal quantile algorithms proposed in the paper are fascinating. It is a very interesting turn in the sketching literature and Qiang and Muthu’s creativity really comes across. I am very interested in getting some real world uses out of this sketch and am excited to see what other applications we (and Qiang!) can think of.  Many thanks to MuthuQiang Ma and Jeremie Lumbroso for all their help!

### Appendix

• Variability: While the bounds on the accuracy of these algorithms seem wide to me, clearly in real world experiments we see much better performance than the guaranteed bounds in the paper. In fact, the error bounds in the paper are dependent on the size of the stream and not fixed.
• Size of step: A potential gotcha is the size of the step variable. Depending on your update function it indeed seems possible for this value to get below -MAXINT. Clearly a real implementation would need some error checking.
• Cold Start: One more gotcha is that you have no real way of knowing when you are near the quantile in question. For instance, starting your estimate at zero and measuring a true median which is 100,000,000,000 will take a long time to “catch up” to this value. There are a few ways to limit this, some of which are discussed in the paper. One way is to try and make a sane guess up front (especially if you know something about the distribution) and another is to start your guess with the value from the last counter. For instance, suppose you are in a monitoring situation and you are computing means over the course of a day. It would make sense to start tomorrow’s estimate with yesterdays final value.
• Accuracy:  And, lastly, there is some interesting dependence on “atomicity”. Meaning, the estimates in some sense depend on how “large” the actual values are. In both, your minimum step size is 1. If my median in the stream is, say, 6 then this “atomic” update of 1 causes high relative fluctuation. It seems in real world examples you would like to balance the size of the thing you are estimating with the speed of approach of the algorithm. This could lead you to estimate latencies in microseconds rather than milliseconds for instance. All of this points to the fact that there are a bunch of real world engineering questions that need to be answered and thought about before you actually go and implement a million frugal sketches throughout your organization.

## Slides and Videos are up for AK Data Science Summit

We’re happy to announce that all the talks (slides and video) from our Data Science Summit are available for free, right here on the blog! You can find a full list of the conference’s content here.

Enjoy!

## Data Science Summit – Update

I don’t think I’m going out on a limb saying that our conference last week was a success. Thanks to everyone who attended and thanks again to all the speakers. Muthu actually beat us to it and wrote up a nice summary. Very kind words from a great guy. For those of you that couldn’t make it (or those that want to relive the fun) we posted the videos and slides. Thanks again to everyone for making it such a success!

Muthu being Muthu during David Woodruff’s fantastic talk on streaming linear algebra

## Open Source Release: js-hll

One of the first things that we wanted to do with HyperLogLog when we first started playing with it was to support and expose it natively in the browser. The thought of allowing users to directly interact with these structures — perform arbitrary unions and intersections on effectively unbounded sets all on the client — was exhilarating to us. We knew it could be done but we simply didn’t have the time.

Fast forward a few years to today. We had finally enough in the meager science/research budget to pick up an intern for a few months and as a side project I tasked him with turning our dream into a reality. Without further ado, we are pleased to announce the open-source release of AK’s HyperLogLog implementation for JavaScript, js-hll. We are releasing this code under the Apache License, Version 2.0 matching our other open source offerings.

We knew that we couldn’t just release a bunch of JavaScript code without allowing you to see it in action — that would be a crime. We passed a few ideas around and the one that kept bubbling to the top was a way to kill two birds with one stone. We wanted something that would showcase what you can do with HLL in the browser and give us a tool for explaining HLLs. It is typical for us to explain how HLL intersections work using a Venn diagram. You draw some overlapping circles with a broder that represents the error and you talk about how if that border is close to or larger than the intersection then you can’t say much about the size of that intersection. This works just ok on a whiteboard but what you really want is to just build a visualization that allows you to select from some sets and see the overlap. Maybe even play with the precision a little bit to see how that changes the result. Well, we did just that!

Click above to interact with the visualization

Note: There’s more interesting math in the error bounds that we haven’t explored. Presenting error bounds on a measurement that cannot mathematically be less than zero is problematic. For instance, if you have a ruler that can only measure to 1/2″ and you measure an object that truly is 1/8″ long you can say “all I know is this object measures under 0.25 inches”. Your object cannot measure less than 0 inches, so you would never say 0 minus some error bound. That is, you DO NOT say 0.0 ± 0.25 inches.  Similarly with set intersections there is no meaning to a negative intersection. We did some digging and just threw our hands up and tossed in what we feel are best practices. In the js-hll code we a) never show negative values and b) we call “spurious” any calculation that results in an answer within 20% of the error bound. If you have a better answer, we would love to hear it!

## Conference Agenda – 9:15 am until 5:30 pm at 111 Minna Gallery

We’ve finalized the agenda for the “Data Science Summit” at 111 Minna. See here for details and the schedule of events. The first talks will start at 10am and we will try to wrap up around 5:30pm. We ask that you start showing up around 9:15 or so to ensure we are ready to start on time. Thanks again to our co-sponsors, Foundation Capital.

Update! We posted the videos and slides for those of you that couldn’t make it (or those that want to relive the fun). Enjoy!

## Open Source Release: js-murmur3-128

As you can imagine from of all of our blog posts about hashing that we hash a lot of things. While the various hashing algorithms may be well-defined, the devil is always in the details especially when working with multiple languages that have different ways of representing numbers. We’re happy to announce the open-source release of AK’s 128bit Murmur3 implementation for JavaScript, js-murmur3-128. We are releasing this code under the Apache License, Version 2.0 matching our other open source offerings.

### Details

The goal of the implementation is to produce a hash value that is equivalent to the C++ and Java (Guava) versions for the same input and it must be usable in the browser. (Full disclosure: we’re still working through some signed/unsigned issues between the C++ and Java/JavaScript versions. The Java and JavaScript versions match exactly.)

### Usage

Java (Guava):

final int seed = 0;
final byte[] bytes = { (byte)0xDE, (byte)0xAD, (byte)0xBE, (byte)0xEF,
(byte)0xFE, (byte)0xED, (byte)0xFA, (byte)0xCE };
.putBytes(bytes)
.hash();
System.err.println(hashCode.asLong());


JavaScript:

var seed = 0;
var rawKey = new ArrayBuffer(8);
var byteView = new Int8Array(rawKey);
byteView[0] = 0xDE; byteView[1] = 0xAD; byteView[2] = 0xBE; byteView[3] = 0xEF;
byteView[4] = 0xFE; byteView[5] = 0xED; byteView[6] = 0xFA; byteView[7] = 0xCE;
console.log(murmur3.hash128(rawKey, seed));