com.stuartsierra.frequencies — com.troy-west/frequencies 0.1.1

Basic statistical computations on frequency maps. A frequency map (freq-map) is a map from observed values to their frequency in a data set.

If the observed values are all integers within a small range, then a frequency map may be exact, such as that returned by clojure.core/frequencies. Floating-point values or a large range of integers can be grouped into 'buckets' as in a histogram: the 'bucket-frequencies' function does this.

You can create your own bucketed frequency map (for example, as part of a larger 'reduce' operation) using the functions 'bucket' and 'recover-bucket-keys'.

bucket clj/s

(bucket bucket-size value)

Returns an integer bucket ID for the observed value. Bucket maps use integers as map keys to avoid possible errors from floating-point arithmetic.

bucket-frequencies clj/s

(bucket-frequencies bucket-size values)

Returns a bucketed frequency map. Keys in the map are values from the input, rounded up to bucket-size. Values in the map are counts of the number of occurances of values less than or equal to their bucket key but greater than the next-lowest bucket key.

mean clj/s

(mean freq-map)

Returns the mean (average) of observed values in a frequency map.

median clj/s

(median freq-map)

Returns the median of the observed values in the frequency map.

percentiles clj/s

(percentiles freq-map percentiles)

Returns a map of percentile values from the frequency map. Argument 'percentiles' is a collection of percentile targets, which will be keys in the returned map. For example, a percentiles argument of [25 50 99.9] would return a map containing the 25th, 50th (median), and 99.9th percentile.

percentiles* clj/s

(percentiles* sorted-freq-map percentiles sample-count)

Like percentiles but the sample-count is provided as an argument instead of computed, and the frequency map must already be sorted.

quantile clj/s

(quantile freq-map k q)

Returns the value which is greater than k/q of the observed values in the frequency map. For example, k=1 q=2 is the median; k=99 q=100 is the 99th percentile. For bucketed frequency maps, returns the nearest bucket.

quantile* clj/s

(quantile* sorted-freq-map k q sample-count)

Like quantile but takes sample-count as an argument. For when you already know the sample-count and don't want to recompute it. Also assumes that the frequency map is already sorted.

recover-bucket-keys clj/s

(recover-bucket-keys bucket-map bucket-size)

Converts the keys of a map from integer bucket IDs to the original value domain. Use this only if you are building up a bucket map yourself with the 'bucket' function; the bucket-frequencies function calls recover-keys automatically.

sample-count clj/s

(sample-count freq-map)

Returns the number of observed values in a frequency map.

stats clj/s

(stats freq-map
       &
       {:keys [percentiles] :or {percentiles [25 50 75 90 95 99 99.9]}})

Returns a map of statistics for the frequency map with the following keys:

:mean, :median, :variance, :stdev, :sum, :sample-count,

:min minimum observed value;

:max maximum observed value;

:percentiles Map of percentile level to observed value. Defaults to quartiles and 90, 95, 99, and 99.9th percentiles. Change the returned percentiles by passing a vector of percentile levels (between 0 and 100) as the option :percentiles.

stdev clj/s

(stdev freq-map)

Returns the standard deviation (square root of the variance) of observed values in a frequency map.

sum clj/s

(sum freq-map)

Returns the sum of all observed values in a frequency map.

values clj/s

(values freq-map)

Returns a lazy sequence of all the observed values, repeating each value the number of times it was observed.

variance clj/s

(variance freq-map)

Returns the variance of observed values in a frequency map.

variance* clj/s

(variance* freq-map mean sample-count)

Like 'variance' but takes the mean and sample count as arguments instead of computing them.