Fast Statistics
A NodeJS library to do statistical analysis of numeric datasets.
introduction
When doing statistical analysis of data, the most common usage pattern is to run multiple statistical methods on the same set of data. Some of these methods use others. For example, to calculate the standard deviation of a dataset, we first need the mean.
Additionally, some methods can be calculated quickly as data is inserted, thereby reducing the number of loops required to run through the data during processing.
Fast stats maintains a running cache of several summary values as data is inserted making final calculation very fast. It trades off a small amount of additional memory usage for a large reduction in execution time.
downsides
The downside of how fast stats operates is that if your datapoints are too large, it may result in numeric overflow causing incorrect results. Fast stats does not attempt to detect or correct for this.
synopsis
var Stats = Stats; var s = ;console;// 4.43
installation
$ npm install faststats
API
faststats is completely synchronous. There are no blocking methods and consequently no callbacks involved. All runtime calls are executed inmemory and are fast.
Configuring the Stats object
The Stats
constructor takes in a configuration object as a parameter. This is a simple keyvalue list that tells
faststats
how to behave under certain conditions.
var s = bucket_precision: 10 ;
The following configuration options are recognised. All of them are optional.

bucket_precision
: [number] Tellsfaststats
to maintain a histogram of your dataset using this parameter as the least count, or precision.This is useful if you have a very large data set, and want to approximate percentile values like the median without having to store the entire dataset in memory. For example, if you had a million time measurements between 0.5 and 1.5 seconds, you could store all million of them, or you could set up 1000 one millisecond buckets and store a count of items in each bucket with a precision of 1 millisecond each. If you reduce (higher values are considered less precise) the precision to 10 milliseconds, the number of buckets reduces from 1000 to 100, taking up less memory overall.
By default,
faststats
will not maintain buckets since it does not know the least count and range of your dataset in advance.This option is required if you need to use the
distribution()
method. 
buckets
: [array of numbers] Tellsfaststats
to maintain a histogram of your dataset using these custom buckets.Each number in the array is the upper limit of a bucket. The lower limit of the first bucket is 0, the lower limit for all other buckets is the upper limit of the previous bucket.
If you use both
bucket_precision
andbuckets
,buckets
takes precedence. 
bucket_extension_interval
: [number] Tellsfaststats
how to extend the predefined buckets if data exceeding the range is added. This is useful to capture data above your range, in multiple buckets, but with low precision so you do not end up with a large number of empty buckets.By default this is not defined, so buckets will not be extended and all data beyond the end range will end up in the last bucket.

store_data
: [boolean] Tellsfaststats
not to store actual data values. This is useful to reduce memory utilisation for large datasets, however it comes with a few caveats. You can no longer get an exact median or other percentile value out of your dataset, however you could
use bucketing (see
bucket_precision
above) to get an approximate percentile value.  You can no longer run an exact
iqr
filter or aband_pass
filter on the data, however you could use bucketing to get an approximate filtered object.  You can no longer get at the entire dataset or remove data from the dataset.
The mean, standard deviation and margin of error calculations are unaffected by this parameter. If you use bucketing, and only care about the mean, standard deviation and margin of error or an approximate median or percentile value, set this option to false.
By default,
store_data
istrue
.  You can no longer get an exact median or other percentile value out of your dataset, however you could
use bucketing (see

sampling
: [boolean] Tellsfaststats
whether the data you pass in is a sample (true
) or the entire (false
default) population.The standard deviation algorithm differs for populations v/s samples.
Getting data in and out
Initialising and adding data
The Stats
object looks a lot like an array in the way you add and remove data to its ends, however there is
no direct access to individual elements. Data is added to the object using the push()
and unshift()
methods.
All values must be numbers and behaviour is undefined if they are not.
The push()
method takes in a list of values that will be added to the end of the current list and
the unshift()
method takes in a list of values that will be added to the beginning of the list.
Instead of passing in multiple parameters, you can also pass in an array of numbers as the first parameter.
The following are equivalent.
var s1 s2 s3 s4;s1 = ; s2 = ; s3 = ;s3; s4 = ;s4;s4; assert;assert;assert;
Note that we use the toFixed()
method of the Number class when comparing numbers. Remember that even if you
pass in integers, values like the arithmetic mean, standard deviation and median can sometimes be floating point
numbers, and two floating point numbers may not necessarily be equal to the last decimal point. The toFixed()
method is useful to restrict how precise we want our comparison to be. Be aware that it returns a string though.
faststats
does not use the toFixed()
method internally.
The push()
and unshift()
methods return the this
object.
Removing data
If you need to remove data from a Stats
object, use the pop()
and shift()
methods. Their semantics are the
same as the pop()
and shift()
methods of Arrays.
var a = s1;assert; var b = s2;assert; assert;assert;assert;
Clearing all data
The reset()
method clears out all data.
s4;assert;
The reset()
method returns a reference to the object, so you can chain methods.
Making a copy
The copy()
method returns a copy of the current Stats object.
s4 = s3; assert;
Additionally, the copy()
method can create a new Stats
object with a different configuration.
This is most useful if you need to change bucket sizes or precision. Simply pass the new config
object as a parameter to the copy()
method:
s4 = s3;
Getting the raw data
The data
instance member returns an array of the raw numbers stored in the current Stats object.
If store_data
is false
, it returns undefined
.
data; // array of numbersstore_data: falsedata; // undefined
Summaries & Averages
The term Average is overloaded in Statistics. It relates to a summary of a data set, but says nothing about how
we arrived at that summary. There are many ways to summarise data, including the arithmetic mean, geometric mean,
harmonic mean, median, mode and more. faststats
implements the Arithmetic Mean, the Geometric Mean and the Median.
It also implements a percentile method to get at any percentile of the data.
Arithmetic Mean
The arithmetic mean is calculated as the sum of all data points divided by the number of data points. This is useful
for data sets that are fairly uniform, following a linear or binomial distribution. Use the amean()
method or the `μ()`
method to get at it:
var a = s1;assert; // remember we popped out the last item of `s1` above.
Geometric Mean
The geometric mean is the n
th root of the product of all data points where n is the number of data points. This is useful
for data sets that follow an exponential or lognormal distribution. Use the gmean()
method to get at it:
var a = s1;assert;
Median
The median is the middle point of the dataset when sorted in ascending order. This is useful if your dataset has a lot of
outliers and noise that would not normally be found in a complete population. Use the median()
method to get at it:
var a = s1;assert;
If your data set contains an odd number of points, the median will be the middle point. If it contains an even number of points, then the median will be the arithmetic mean of the two middle points.
If your Stats object is configured to use buckets and has store_data
set to false, then the median will be an approximation
of the actual median.
Any Percentile
You can also get at any percentile value within the data. Use the percentile()
method to get at this data. The
percentile()
method takes in a single argument. This is a number between 0 and 100 (both inclusive) that specifies
which percentile point you want.
var p95 = s1;var m = s1;var q1 = s1; assert;assert;assert;
Passing in 50 as an argument will return the median, while 25 and 75 will return the first and third quartiles respectively. These three special values may be arithmetic means of two other values within the set. All other arguments will return a number from the data set.
If your Stats object is configured to use buckets and has store_data
set to false, then the percentile value returned will
be an approximation of the actual percentile based on the configured bucket_precision
or buckets
.
Range
The range()
method tells you the minimum and maximum values of your data set. It returns an array of two values. The
first is the lower bound and the second is the upper bound.
var r = s1; assert;assert;assert;
Distribution
The distribution()
method tells you how your data is distributed. You need to set the bucket_precision
or buckets
configuration options if you plan on using this method. It will then split your data into buckets based on the value of
bucket_precision
or buckets
and tell you how many data points fall into each bucket. You can use this to plot a
histogram of your data, or to compare it to commonly known distribution functions.
The return value is a sparse array of buckets with counts of datapoints per bucket. To save on memory, any empty buckets are undefined. You should treat an undefined bucket as if it had 0 datapoints.
A bucket structure looks like this:
bucket: <bucket midpoint> range: <bucket low> <bucket high> count: <number of datapoints>
Note that the upper bound of the range
is open, ie, the range does not include the upper bound.
var s7 = bucket_precision: 10; // Populate s7 with sequence of squares from 010// 0 1 4 9 16 25 36 49 64 81 100forvar i=0; i<=10; i++ s7; // distribution should be [4, 1, 1, 1, 1, 0, 1, 0, 1, 0, 1]// but 0s are undefined to save on memoryvar d=s7; // length should be one more than (maxmin)/bucket_precisionassert; d;
Using custom buckets instead:
var assert = Stats = Stats; var s1 = buckets: 1 2 3 5 8 13;forvar i=0; i<20; i++ s1; var d = s1; d;
Data Accuracy
There are various statistical values that tell you how accurate or uniform your data is. faststats
implements
the Arithmetic Standard Deviation, Geometric Standard Deviation and 95% Confidence Interval Margin of Error.
Arithmetic Standard Deviation
Also commonly just called the Standard Deviation, with the symbol σ. This tells you the spread of your data if
it follows a normal (or close to normal) distribution, ie, the bell curve. faststats
is really fast at
calculating the standard deviation of a dataset. Use the stddev()
method or the `σ()` method to get at it.
var sd = s1; assert;
The arithmetic standard deviation is used in conjunction with the arithmetic mean to tell you the spread of your
dataset: [ameanstddev, amean+stddev]
. Note that you could also use 2 or 3 standard deviations for different
spreads.
Geometric Standard Deviation
The geometric standard deviation tells you the spread of your data if it follows a lognormal or exponential distribution.
Use the gstddev()
method to get at it.
var gsd = s1; assert;
The geometric standard deviation is used in conjunction with the geometric mean to tell you the spread of your
dataset: [gmean/gstddev, gmean*gstddev]
. Note that this range is not symmetric around the geometric mean.
95% Confidence Margin of Error
The Margin of Error value tells you the range within which the real arithmetic mean of the population is likely to
be with 95% confidence. Use the moe()
method to get at it.
var moe = s1; assert;
This value suggests that we are 95% certain that the real mean of the population is within 2.60 of the calculated arithmetic mean of 4.67. We could use this to find out the percent error in our sample. In this case there is a 55.71% error.
The margin of error is inversely proportional to the square root of the number of data points, so increasing the size of your sample will reduce the margin of error. It is good to strive for a margin of error of less than 5%.
Data filtering
When dealing with statistical samples, it may be necessary to filter the dataset to get rid of outliers. Sometimes an outlier is fairly obvious, and you can specify an upper and lower limit for it. At other times, outliers are only apparent when looking at the rest of the dataset. InterQuartileRange filtering is useful to filter out these kinds of data sets.
Note that if your Stats object is configured to use buckets and has store_data
set to false, then all filtering
will be done on an approximation of the data based on the configured value of bucket_precision
. For example,
if you have a set of numbers from 1100 with bucket_precision
set to 1, then filtering the dataset between 55
and 85 will get you a dataset between 55 and 85. If instead, bucket_precision
is set to 10, then the filtered
dataset will approximately range from 50 to 90. Note, however, that the range()
method will attempt to match as
closely as possible the real range.
Bandpass filtering
The band_pass()
filter method returns a new Stats
object with all its data points within the specified range.
This method takes in three arguments. The first is the lower bound of the range, the second is the upper bound
of the range. Both these arguments are required.
The third argument specifies whether the range is open or closed. An open range does not include the upper and
lower bounds while a closed range includes them. If not specified (or set to false
), the range is closed. If
set to true
the range is open.
var s5 = s1;var r = s5; assert;assert; s5 = s1;r = s5; assert;assert;
Band pass filtering should be used if the range for your data is rigid and never changes.
IQR Filtering
IQR, or Inter Quartile Range filtering filters data based on the spread of the data. It is much more adaptive to
changes in data ranges. Use the iqr()
method to IQR filter a dataset. The iqr()
method does not accept
any arguments.
var s6 = s1;r = s6; assert;assert;
In some cases, IQR filtering may not filter out anything. This can happen if the acceptable range is wider than the bounds of your dataset.
References
Wikipedia is a great place to get information about Statistical functions.
Copyright
faststats
is Copyright 2011 Philip Tellis philip@bluesmoon.info and the latest version of the code is
available at https://github.com/bluesmoon/nodefaststats
License
Apache 2.0. See the LICENSE file for details.