ajet/df-research
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
main
df-research/tech.ml.dataset/docs/walkthrough.html
ajet bdf064f54d init research
2026-02-08 11:20:43 -10:00

740 lines
47 KiB
HTML
Vendored
Raw Permalink Blame History

<!DOCTYPE html PUBLIC ""
"">
<html><head><meta charset="UTF-8" /><title>tech.ml.dataset Walkthrough</title><script async="true" src="https://www.googletagmanager.com/gtag/js?id=G-RGTB4J7LGP"></script><script>window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-95TVFC1FEB');</script><link rel="stylesheet" type="text/css" href="css/default.css" /><link rel="stylesheet" type="text/css" href="highlight/solarized-light.css" /><script type="text/javascript" src="highlight/highlight.min.js"></script><script type="text/javascript" src="js/jquery.min.js"></script><script type="text/javascript" src="js/page_effects.js"></script><script>hljs.initHighlightingOnLoad();</script></head><body><div id="header"><h2>Generated by <a href="https://github.com/weavejester/codox">Codox</a> with <a href="https://github.com/xsc/codox-theme-rdash">RDash UI</a> theme</h2><h1><a href="index.html"><span class="project-title"><span class="project-name">TMD</span> <span class="project-version">7.000-beta-23</span></span></a></h1></div><div class="sidebar primary"><h3 class="no-link"><span class="inner">Project</span></h3><ul class="index-link"><li class="depth-1 "><a href="index.html"><div class="inner">Index</div></a></li></ul><h3 class="no-link"><span class="inner">Topics</span></h3><ul><li class="depth-1 "><a href="columns-readers-and-datatypes.html"><div class="inner"><span>tech.ml.dataset Columns, Readers, and Datatypes</span></div></a></li><li class="depth-1 "><a href="nippy-serialization-rocks.html"><div class="inner"><span>tech.ml.dataset And nippy</span></div></a></li><li class="depth-1 "><a href="quick-reference.html"><div class="inner"><span>tech.ml.dataset Quick Reference</span></div></a></li><li class="depth-1 "><a href="supported-datatypes.html"><div class="inner"><span>tech.ml.dataset Supported Datatypes</span></div></a></li><li class="depth-1 current"><a href="walkthrough.html"><div class="inner"><span>tech.ml.dataset Walkthrough</span></div></a></li></ul><h3 class="no-link"><span class="inner">Namespaces</span></h3><ul><li class="depth-1"><div class="no-link"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>tech</span></div></div></li><li class="depth-2"><div class="no-link"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>v3</span></div></div></li><li class="depth-3"><a href="tech.v3.dataset.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>dataset</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.categorical.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>categorical</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.clipboard.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>clipboard</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.column.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>column</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.column-filters.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>column-filters</span></div></a></li><li class="depth-4"><div class="no-link"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>io</span></div></div></li><li class="depth-5 branch"><a href="tech.v3.dataset.io.csv.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>csv</span></div></a></li><li class="depth-5 branch"><a href="tech.v3.dataset.io.datetime.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>datetime</span></div></a></li><li class="depth-5 branch"><a href="tech.v3.dataset.io.string-row-parser.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>string-row-parser</span></div></a></li><li class="depth-5"><a href="tech.v3.dataset.io.univocity.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>univocity</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.join.html"><div class="inner"><span class="tree" style="top: -145px;"><span class="top" style="height: 154px;"></span><span class="bottom"></span></span><span>join</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.math.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>math</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.metamorph.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>metamorph</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.modelling.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>modelling</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.neanderthal.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>neanderthal</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.print.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>print</span></div></a></li><li class="depth-4"><a href="tech.v3.dataset.reductions.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>reductions</span></div></a></li><li class="depth-5"><a href="tech.v3.dataset.reductions.apache-data-sketch.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>apache-data-sketch</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.rolling.html"><div class="inner"><span class="tree" style="top: -52px;"><span class="top" style="height: 61px;"></span><span class="bottom"></span></span><span>rolling</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.set.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>set</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.dataset.tensor.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>tensor</span></div></a></li><li class="depth-4"><a href="tech.v3.dataset.zip.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>zip</span></div></a></li><li class="depth-3"><div class="no-link"><div class="inner"><span class="tree" style="top: -672px;"><span class="top" style="height: 681px;"></span><span class="bottom"></span></span><span>libs</span></div></div></li><li class="depth-4 branch"><a href="tech.v3.libs.arrow.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>arrow</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.libs.fastexcel.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>fastexcel</span></div></a></li><li class="depth-4"><div class="no-link"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>guava</span></div></div></li><li class="depth-5"><a href="tech.v3.libs.guava.cache.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>cache</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.libs.parquet.html"><div class="inner"><span class="tree" style="top: -52px;"><span class="top" style="height: 61px;"></span><span class="bottom"></span></span><span>parquet</span></div></a></li><li class="depth-4 branch"><a href="tech.v3.libs.poi.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>poi</span></div></a></li><li class="depth-4"><div class="no-link"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>smile</span></div></div></li><li class="depth-5"><a href="tech.v3.libs.smile.data.html"><div class="inner"><span class="tree"><span class="top"></span><span class="bottom"></span></span><span>data</span></div></a></li><li class="depth-4"><a href="tech.v3.libs.tribuo.html"><div class="inner"><span class="tree" style="top: -52px;"><span class="top" style="height: 61px;"></span><span class="bottom"></span></span><span>tribuo</span></div></a></li></ul></div><div class="document" id="content"><div class="doc"><div class="markdown"><h1>tech.ml.dataset Walkthrough</h1>
<p>Let's take a moment to walkthrough the <code>tech.ml.dataset</code> system. This system was built
over the course of a few months in order to make working with columnar data easier
in the same manner as one would work with <code>data.table</code> in R or <code>pandas</code> in Python. While
it takes design inspiration from these sources it does not strive to be a copy in any
way but rather an extension to the core Clojure language that is built for good
performance when processing datasets of realistic sizes which in our case means
millions of rows and tens of columns.</p>
<h2>High Level Design</h2>
<p>Logically, a dataset is a map of column name to column data. Column data is typed
so for instance you may have a column of 16 bit integers or 64 bit floating point
numbers. Column names may be any java object and column values may be of the
tech.datatype primitive, datetime, or objects. Data is stored contiguously in jvm
arrays while missing values are indicated with bitsets.</p>
<p>Given this definition, the intention is to allow more or less normal flows familiar
to most Clojure programmers:</p>
<ol>
<li>Dataset creation in the form of csv,tsv (and gzipped varieties of these), maps of
column name to column values, and arbitrary sequences of maps.</li>
<li>Pretty printing of datasets and, to a lesser extent, columns. Simple selection of
a given column and various functions describing the details of a column.</li>
<li>Access to the values in a column including eliding or erroring on missing values.</li>
<li>Select subrect of dataset defined by a sequence of columns and some sequence of
indexes.</li>
<li><code>sort-by</code>, <code>filter</code>, <code>group-by</code> are modified operations that operate on a
logical sequence of maps and an arbitrary function but return a new dataset.</li>
<li>Efficient elementwise operations such as linear combinations of columns.</li>
<li>Statistical and ml-based analysis of some subset of columns either on their own
or as they relate to another <code>target</code> column.</li>
<li>Conversion of the dataset to sequences of maps, sequences of persistent vectors, and
rowwise sequences of java arrays of a chosen primitive datatype.</li>
</ol>
<h2>Dataset Creation</h2>
<h4>-&gt;dataset -&gt;&gt;dataset</h4>
<p>Dataset creation can happen in many ways. For data in csv, tsv, or sequence of maps
format there are two functions that differ in where the data is passed in, <code>-&gt;dataset</code>
and <code>-&gt;&gt;dataset</code>. These functions several arguments:</p>
<ul>
<li>A <code>String</code> or <code>InputStream</code> will be interpreted as a file (or gzipped file if it
ends with .gz) of tsv or csv data. The system will attempt to autodetect if this
is csv or tsv and then has some extensive engineering put into column datatype
detection mechanisms which can be overridden.</li>
<li>A sequence of maps may be passed in in which case the first N maps are scanned in
order to derive the column datatypes before the actual columns are created.</li>
</ul>
<pre><code class="language-clojure">user&gt; (require '[tech.v3.dataset :as ds])
nil
user&gt; (ds/-&gt;dataset [{:a 1 :b 2} {:a 2 :c 3}])
_unnamed [2 3]:
| :a | :b | :c |
|----|----|----|
| 1 | 2 | |
| 2 | | 3 |
</code></pre>
<h4>CSV/TSV/MAPSEQ/XLS/XLSX Parsing Options</h4>
<p>It is important to note that there are many options for parsing files.
A few important ones are column allowlist/blocklists, num records,
and ways to specify exactly how to parse the string data:</p>
<ul>
<li><a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var--.3Edataset">https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var--.3Edataset</a></li>
</ul>
<pre><code class="language-clojure">
user&gt; (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz"
{:column-whitelist ["SalePrice" "1stFlrSF" "2ndFlrSF"]
:n-records 5})
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 3]:
| SalePrice | 1stFlrSF | 2ndFlrSF |
|-----------|----------|----------|
| 208500 | 856 | 854 |
| 181500 | 1262 | 0 |
| 223500 | 920 | 866 |
| 140000 | 961 | 756 |
| 250000 | 1145 | 1053 |
user&gt; (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz"
{:column-whitelist ["SalePrice" "1stFlrSF" "2ndFlrSF"]
:n-records 5
:parser-fn :float32})
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 3]:
| SalePrice | 1stFlrSF | 2ndFlrSF |
|-----------|----------|----------|
| 208500.0 | 856.0 | 854.0 |
| 181500.0 | 1262.0 | 0.0 |
| 223500.0 | 920.0 | 866.0 |
| 140000.0 | 961.0 | 756.0 |
| 250000.0 | 1145.0 | 1053.0 |
user&gt; (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz"
{:column-whitelist ["SalePrice" "1stFlrSF" "2ndFlrSF"]
:n-records 5
:parser-fn {"SalePrice" :float32}})
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [4 3]:
| SalePrice | 1stFlrSF | 2ndFlrSF |
|-----------|----------|----------|
| 208500.0 | 856 | 854 |
| 181500.0 | 1262 | 0 |
| 223500.0 | 920 | 866 |
| 140000.0 | 961 | 756 |
| 250000.0 | 1145 | 1053 |
</code></pre>
<p>You can also supply a tuple of <code>[datatype parse-fn]</code> if you have a specific
datatype and parse function you want to use. For datetime types <code>parse-fn</code>
can additionally be a DateTimeFormat format string or a DateTimeFormat object:</p>
<pre><code class="language-clojure">user&gt; (require '[tech.v3.libs.fastexcel])
nil
user&gt; (def data (ds/head (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx")))
#'user/data
user&gt; data
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx [5 8]:
| column-0 | First Name | Last Name | Gender | Country | Age | Date | Id |
|----------|------------|-----------|--------|---------------|------|------------|--------|
| 1.0 | Dulce | Abril | Female | United States | 32.0 | 15/10/2017 | 1562.0 |
| 2.0 | Mara | Hashimoto | Female | Great Britain | 25.0 | 16/08/2016 | 1582.0 |
| 3.0 | Philip | Gent | Male | France | 36.0 | 21/05/2015 | 2587.0 |
| 4.0 | Kathleen | Hanner | Female | United States | 25.0 | 15/10/2017 | 3549.0 |
| 5.0 | Nereida | Magwood | Female | United States | 58.0 | 16/08/2016 | 2468.0 |
user&gt; ;; Note the Date actually didn't parse out because it is dd/MM/yyyy format:
user&gt; (meta (data "Date"))
{:categorical? true, :name "Date", :datatype :string, :n-elems 5}
user&gt; (def data (ds/head (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx"
{:parser-fn {"Date" [:local-date "dd/MM/yyyy"]}})))
#'user/data
user&gt; data
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx [5 8]:
| column-0 | First Name | Last Name | Gender | Country | Age | Date | Id |
|----------|------------|-----------|--------|---------------|------|------------|--------|
| 1.0 | Dulce | Abril | Female | United States | 32.0 | 2017-10-15 | 1562.0 |
| 2.0 | Mara | Hashimoto | Female | Great Britain | 25.0 | 2016-08-16 | 1582.0 |
| 3.0 | Philip | Gent | Male | France | 36.0 | 2015-05-21 | 2587.0 |
| 4.0 | Kathleen | Hanner | Female | United States | 25.0 | 2017-10-15 | 3549.0 |
| 5.0 | Nereida | Magwood | Female | United States | 58.0 | 2016-08-16 | 2468.0 |
user&gt; (meta (data "Date"))
{:name "Date", :datatype :local-date, :n-elems 5}
user&gt; (nth (data "Date") 0)
#object[java.time.LocalDate 0x6c88bf34 "2017-10-15"]
user&gt; (def data (ds/head (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx"
{:parser-fn {"Date" [:local-date "dd/MM/yyyy"]
"Id" :int32
"column-0" :int32
"Age" :int16}})))
#'user/data
user&gt; data
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/file_example_XLSX_1000.xlsx [5 8]:
| column-0 | First Name | Last Name | Gender | Country | Age | Date | Id |
|----------|------------|-----------|--------|---------------|-----|------------|------|
| 1 | Dulce | Abril | Female | United States | 32 | 2017-10-15 | 1562 |
| 2 | Mara | Hashimoto | Female | Great Britain | 25 | 2016-08-16 | 1582 |
| 3 | Philip | Gent | Male | France | 36 | 2015-05-21 | 2587 |
| 4 | Kathleen | Hanner | Female | United States | 25 | 2017-10-15 | 3549 |
| 5 | Nereida | Magwood | Female | United States | 58 | 2016-08-16 | 2468 |
</code></pre>
<h4>Map Of Columns Format</h4>
<p>Given a map of name-&gt;column data produce a new dataset. If column data is untyped
(like a persistent vector) then the column datatype is either string or double,
dependent upon the first entry of the column data sequence.</p>
<p>If the column data is one of the object numeric primitive types, so
<code>Float</code> as opposed to <code>float</code>, then missing elements will be marked as
missing and the default empty-value will be used in the primitive storage.</p>
<pre><code class="language-clojure">
user&gt; (ds/-&gt;dataset {:age [1 2 3 4 5] :name ["a" "b" "c" "d" "e"]})
_unnamed [5 2]:
| :age | :name |
|------|-------|
| 1 | a |
| 2 | b |
| 3 | c |
| 4 | d |
| 5 | e |
</code></pre>
<h2>Printing</h2>
<p>Printing out datasets comes in several flavors. Datasets support multiline printing:</p>
<pre><code class="language-clojure">user&gt; (require '[tech.v3.tensor :as dtt])
nil
user&gt; (def test-tens (dtt/-&gt;tensor (partition 3 (range 9))))
#'user/test-tens
user&gt; (ds/-&gt;dataset [{:a 1 :b test-tens}{:a 2 :b test-tens}])
_unnamed [2 2]:
| :a | :b |
|----|------------------------------|
| 1 | #tech.v3.tensor&lt;object&gt;[3 3] |
| | [[0 1 2] |
| | [3 4 5] |
| | [6 7 8]] |
| 2 | #tech.v3.tensor&lt;object&gt;[3 3] |
| | [[0 1 2] |
| | [3 4 5] |
| | [6 7 8]] |
</code></pre>
<p>You can provide options to control printing via the metadata of the dataset:</p>
<pre><code class="language-clojure">user&gt; (def tens-ds *1)
#'user/tens-ds
user&gt; (with-meta tens-ds
(assoc (meta tens-ds)
:print-line-policy :single))
_unnamed [2 2]:
| :a | :b |
|----|------------------------------|
| 1 | #tech.v3.tensor&lt;object&gt;[3 3] |
| 2 | #tech.v3.tensor&lt;object&gt;[3 3] |
</code></pre>
<p>This is especially useful when dealing with new datasets that may have large amounts
of per-column data:</p>
<pre><code class="language-clojure">user&gt; (def events-ds (-&gt; (ds/-&gt;dataset "https://api.github.com/events"
{:key-fn keyword
:file-type :json})
(vary-meta assoc :print-line-policy :single
:print-column-max-width 25)))
#'user/events-ds
user&gt; (ds/head events-ds)
https://api.github.com/events [5 8]:
| :id | :type | :actor | :repo | :payload | :public | :created_at | :org |
|-------------|------------------------|----------------|-----------------|-----------------------|---------|----------------------|----------------|
| 13911736787 | PushEvent | {:id 29139614, | {:id 253259114, | {:push_id 5888739305, | true | 2020-10-20T17:49:36Z | |
| 13911736794 | IssuesEvent | {:id 47793873, | {:id 240806054, | {:action "opened", | true | 2020-10-20T17:49:36Z | {:id 61098177, |
| 13911736759 | PushEvent | {:id 71535163, | {:id 304746399, | {:push_id 5888739282, | true | 2020-10-20T17:49:36Z | |
| 13911736795 | PullRequestReviewEvent | {:id 47063667, | {:id 305218173, | {:action "created", | true | 2020-10-20T17:49:36Z | |
| 13911736760 | PushEvent | {:id 22623307, | {:id 287289752, | {:push_id 5888739280, | true | 2020-10-20T17:49:36Z | |
</code></pre>
<p>The full list of possible options is provided in the documentation for <a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.print.html">dataset-data-&gt;str</a>.</p>
<h2>Basic Dataset Manipulation</h2>
<p>Dataset are implementations of <code>clojure.lang.IPersistentMap</code>. They strictly
respect column ordering, however, unlike persistent maps.</p>
<pre><code class="language-clojure">
user&gt; (def new-ds (ds/-&gt;dataset [{:a 1 :b 2} {:a 2 :c 3}]))
#'user/new-ds
user&gt; new-ds
_unnamed [2 3]:
| :a | :b | :c |
|----|----|----|
| 1 | 2 | |
| 2 | | 3 |
user&gt; (first new-ds)
[:a #tech.v3.dataset.column&lt;int64&gt;[2]
:a
[1, 2, ]]
user&gt; (new-ds :c)
#tech.v3.dataset.column&lt;int64&gt;[2]
:c
[, 3, ]
user&gt; (ds/missing (new-ds :b))
{1}
user&gt; (ds/missing (new-ds :c))
{0}
</code></pre>
<p>It is safe to print out very large columns. The system will only print out the first
20 or values. In this way it can be useful to get a feel for the data in a particular
column.</p>
<h2>Access To Column Values</h2>
<p>Columns implement <code>clojure.lang.Indexed</code> (provides nth) and also implement
<code>clojure.lang.IFn</code> in the same manner as persistent vectors.</p>
<pre><code class="language-clojure">user&gt; (ds/-&gt;dataset {:age [1 2 3 4 5]
:name ["a" "b" "c" "d" "e"]})
_unnamed [5 2]:
| :age | :name |
|------|-------|
| 1 | a |
| 2 | b |
| 3 | c |
| 4 | d |
| 5 | e |
user&gt; (def nameage *1)
#'user/nameage
user&gt; (require '[tech.v3.datatype :as dtype])
nil
user&gt; (dtype/-&gt;array-copy (nameage :age))
[1, 2, 3, 4, 5]
user&gt; (type *1)
[J
user&gt; (def namecol (nameage :age))
#'user/namecol
user&gt; (namecol 0)
1
user&gt; (namecol 1)
2
</code></pre>
<p>In the same vein, you can access entire rows of the dataset as a reader that converts
the data either into a persistent vector in the same column-order as the dataset or
a sequence of maps with each entry named. This type of conversion does not include
any mapping to or from labelled values so as such represented the dataset as it is
stored in memory:</p>
<pre><code class="language-clojure">user&gt; (ds/rowvecs nameage)
[[1 "a"] [2 "b"] [3 "c"] [4 "d"] [5 "e"]]
user&gt; (ds/rows nameage)
[{:name "a", :age 1} {:name "b", :age 2} {:name "c", :age 3} {:name "d", :age 4} {:name "e", :age 5}]
</code></pre>
<h2>Subrect Selection</h2>
<p>The dataset system offers two methods to select subrects of information from the
dataset. This results in a new dataset.</p>
<pre><code class="language-clojure">(def ames-ds (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz"))
#'user/ames-ds
user&gt; (ds/column-names ames-ds)
("Id"
"MSSubClass"
"MSZoning"
"LotFrontage"
...)
user&gt; (ames-ds "KitchenQual")
#tech.ml.dataset.column&lt;string&gt;[1460]
KitchenQual
[Gd, TA, Gd, Gd, Gd, TA, Gd, TA, TA, TA, TA, Ex, TA, Gd, TA, TA, TA, TA, Gd, TA, ...]
user&gt; (ames-ds "SalePrice")
#tech.ml.dataset.column&lt;int32&gt;[1460]
SalePrice
[208500, 181500, 223500, 140000, 250000, 143000, 307000, 200000, 129900, 118000, 129500, 345000, 144000, 279500, 157000, 132000, 149000, 90000, 159000, 139000, ...]
user&gt; (ds/select ames-ds ["KitchenQual" "SalePrice"] [1 3 5 7 9])
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| KitchenQual | SalePrice |
|-------------|-----------|
| TA | 181500 |
| Gd | 140000 |
| TA | 143000 |
| TA | 200000 |
| TA | 118000 |
user&gt; (ds/head (ds/select-columns ames-ds ["KitchenQual" "SalePrice"]))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| KitchenQual | SalePrice |
|-------------|-----------|
| Gd | 208500 |
| TA | 181500 |
| Gd | 223500 |
| Gd | 140000 |
| Gd | 250000 |
</code></pre>
<h2>Add, Remove, Update</h2>
<pre><code class="language-clojure">user&gt; (require '[tech.v3.datatype.functional :as dfn])
nil
user&gt; (def small-ames (ds/head (ds/select-columns ames-ds ["KitchenQual" "SalePrice"])))
#'user/small-ames
user&gt; small-ames
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| KitchenQual | SalePrice |
|-------------|-----------|
| Gd | 208500 |
| TA | 181500 |
| Gd | 223500 |
| Gd | 140000 |
| Gd | 250000 |
user&gt; (assoc small-ames "SalePriceLog" (dfn/log (small-ames "SalePrice")))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 3]:
| KitchenQual | SalePrice | SalePriceLog |
|-------------|-----------|--------------|
| Gd | 208500 | 12.24769432 |
| TA | 181500 | 12.10901093 |
| Gd | 223500 | 12.31716669 |
| Gd | 140000 | 11.84939770 |
| Gd | 250000 | 12.42921620 |
user&gt; (assoc small-ames "Range" (range) "Constant-Col" :a)
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 4]:
| KitchenQual | SalePrice | Range | Constant-Col |
|-------------|-----------|-------|--------------|
| Gd | 208500 | 0 | :a |
| TA | 181500 | 1 | :a |
| Gd | 223500 | 2 | :a |
| Gd | 140000 | 3 | :a |
| Gd | 250000 | 4 | :a |
user&gt; (dissoc small-ames "KitchenQual")
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 1]:
| SalePrice |
|-----------|
| 208500 |
| 181500 |
| 223500 |
| 140000 |
| 250000 |
</code></pre>
<h2>Sort-by, Filter, Group-by</h2>
<p>These functions do conceptually the same thing but the dataset is the <em>first</em> argument
so when we build large pipelines of dataset functionaly we don't have to switch the
argument orders. They have per-column versions that are more efficient than the
whole-dataset versions.</p>
<p>The whole-dataset version pass in each row as a map so it is conceptually similar to doing
something like <code>(-&gt;&gt; (ds/mapseq-reader ds) (clojure.core/filter pred))</code>.</p>
<pre><code class="language-clojure">user&gt; (-&gt; ames-ds
(ds/filter-column "SalePrice" #(&lt; 30000 %))
(ds/select ["SalePrice" "KitchenQual"] (range 5)))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| SalePrice | KitchenQual |
|-----------|-------------|
| 208500 | Gd |
| 181500 | TA |
| 223500 | Gd |
| 140000 | Gd |
| 250000 | Gd |
user&gt; ;;Using full dataset version of filter
user&gt; (-&gt; ames-ds
(ds/filter #(&lt; 30000 (get % "SalePrice")))
(ds/select ["SalePrice" "KitchenQual"] (range 5)))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| SalePrice | KitchenQual |
|-----------|-------------|
| 208500 | Gd |
| 181500 | TA |
| 223500 | Gd |
| 140000 | Gd |
| 250000 | Gd |
user&gt; (-&gt; (ds/sort-by-column ames-ds "SalePrice")
(ds/select ["SalePrice" "KitchenQual"] (range 5)))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| SalePrice | KitchenQual |
|-----------|-------------|
| 34900 | TA |
| 35311 | TA |
| 37900 | TA |
| 39300 | Fa |
| 40000 | TA |
user&gt; (def group-map (-&gt; (ds/select ames-ds ["SalePrice" "KitchenQual"] (range 20))
(ds/group-by-column "KitchenQual")))
#'user/group-map
user&gt; (keys group-map)
("Ex" "TA" "Gd")
user&gt; (first group-map)
["Ex" Ex [1 2]:
| SalePrice | KitchenQual |
|-----------|-------------|
| 345000 | Ex |
]
</code></pre>
<p>Combining a <code>group-by</code> variant with <code>descriptive-stats</code> can quickly help break down
a dataset as it relates to a categorical value:</p>
<pre><code class="language-clojure">user&gt; (as-&gt; (ds/select-columns ames-ds ["SalePrice" "KitchenQual" "BsmtFinSF1" "GarageArea"]) ds
(ds/group-by-column ds "KitchenQual")
(map (fn [[k v-ds]]
(-&gt; (ds/descriptive-stats v-ds)
(ds/set-dataset-name k))) ds))
(Ex [4 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-------------|-----------|----------|------------|---------|-----------|-------|----------|---------------------|-------------|
| BsmtFinSF1 | :int16 | 100 | 0 | 0.0 | 850.61 | | 5644.0 | 799.3833216 | 2.14350280 |
| GarageArea | :int16 | 100 | 0 | 0.0 | 706.43 | | 1418.0 | 236.2931861 | -0.18707598 |
| KitchenQual | :string | 100 | 0 | | | Ex | | | |
| SalePrice | :int32 | 100 | 0 | 86000.0 | 328554.67 | | 755000.0 | 120862.9425733 | 0.93681387 |
Fa [4 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-------------|-----------|----------|------------|---------|-----------------|-------|----------|---------------------|------------|
| BsmtFinSF1 | :int16 | 39 | 0 | 0.0 | 136.51282051 | | 932.0 | 209.11654668 | 1.97463203 |
| GarageArea | :int16 | 39 | 0 | 0.0 | 214.56410256 | | 672.0 | 201.93443371 | 0.42348196 |
| KitchenQual | :string | 39 | 0 | | | Fa | | | |
| SalePrice | :int32 | 39 | 0 | 39300.0 | 105565.20512821 | | 200000.0 | 36004.25403680 | 0.24228279 |
TA [4 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-------------|-----------|----------|------------|---------|-----------------|-------|----------|---------------------|------------|
| BsmtFinSF1 | :int16 | 735 | 0 | 0.0 | 394.33741497 | | 1880.0 | 360.21459000 | 0.62751158 |
| GarageArea | :int16 | 735 | 0 | 0.0 | 394.24081633 | | 1356.0 | 187.55679385 | 0.17455203 |
| KitchenQual | :string | 735 | 0 | | | TA | | | |
| SalePrice | :int32 | 735 | 0 | 34900.0 | 139962.51156463 | | 375000.0 | 38896.28033636 | 0.99865115 |
Gd [4 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-------------|-----------|----------|------------|---------|-----------------|-------|----------|---------------------|------------|
| BsmtFinSF1 | :int16 | 586 | 0 | 0.0 | 456.46928328 | | 1810.0 | 455.20910936 | 0.59724411 |
| GarageArea | :int16 | 586 | 0 | 0.0 | 549.10068259 | | 1069.0 | 174.38742143 | 0.22683853 |
| KitchenQual | :string | 586 | 0 | | | Gd | | | |
| SalePrice | :int32 | 586 | 0 | 79000.0 | 212116.02389078 | | 625000.0 | 64020.17670212 | 1.18880409 |
)
</code></pre>
<h4>Rowwise Operations</h4>
<p>Datasets have efficient parallelized mechanisms of presenting data for rowwise map and mapcat
operations. The maps passed into the mapping functions are maps that lazily read
only the required data from the underlying dataset. The returned maps will be
scanned to gather datatype and missing information. Columns derived from the mapping
operation will overwrite columns in the original dataset.</p>
<p>The mapping operations are run in parallel using a primitive named <code>pmap-ds</code> and the resulting
datasets can either be returned in a sequence or combined into a single larger dataset.</p>
<ul>
<li><a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var-rows">rows</a></li>
<li><a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var-rowvecs">rowvecs</a></li>
<li><a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var-row-map">row-map</a></li>
<li><a href="https://techascent.github.io/tech.ml.dataset/tech.v3.dataset.html#var-row-mapcat">row-mapcat</a></li>
</ul>
<h4>Descriptive Stats And GroupBy And DateTime Types</h4>
<p>This is best illustrated by an example:</p>
<pre><code class="language-clojure">user&gt; (def stocks (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv"))
#'user/stocks
user&gt; (ds/select-rows stocks (range 5))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv [5 3]:
| symbol | date | price |
|--------|------------|-------|
| MSFT | 2000-01-01 | 39.81 |
| MSFT | 2000-02-01 | 36.35 |
| MSFT | 2000-03-01 | 43.22 |
| MSFT | 2000-04-01 | 28.37 |
| MSFT | 2000-05-01 | 25.45 |
user&gt; (-&gt;&gt; (ds/group-by-column stocks "symbol")
(map (fn [[k v]] (ds/descriptive-stats v))))
(MSFT: descriptive-stats [3 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-----------|--------------------|----------|------------|------------|------------|-------|------------|---------------------|------------|
| date | :packed-local-date | 123 | 0 | 2000-01-01 | 2005-01-30 | | 2010-03-01 | 9.37554538E+10 | 0.00025335 |
| price | :float64 | 123 | 0 | 15.81 | 24.74 | | 43.22 | 4.30395786E+00 | 1.16559225 |
| symbol | :string | 123 | 0 | | | MSFT | | | |
GOOG: descriptive-stats [3 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-----------|--------------------|----------|------------|------------|------------|-------|------------|---------------------|-------------|
| date | :packed-local-date | 68 | 0 | 2004-08-01 | 2007-05-17 | | 2010-03-01 | 5.20003989E+10 | 0.00094625 |
| price | :float64 | 68 | 0 | 102.4 | 415.9 | | 707.0 | 1.35069851E+02 | -0.22776524 |
| symbol | :string | 68 | 0 | | | GOOG | | | |
AAPL: descriptive-stats [3 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-----------|--------------------|----------|------------|------------|------------|-------|------------|---------------------|------------|
| date | :packed-local-date | 123 | 0 | 2000-01-01 | 2005-01-30 | | 2010-03-01 | 9.37554538E+10 | 0.00025335 |
| price | :float64 | 123 | 0 | 7.070 | 64.73 | | 223.0 | 6.31237823E+01 | 0.93215285 |
| symbol | :string | 123 | 0 | | | AAPL | | | |
IBM: descriptive-stats [3 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-----------|--------------------|----------|------------|------------|------------|-------|------------|---------------------|------------|
| date | :packed-local-date | 123 | 0 | 2000-01-01 | 2005-01-30 | | 2010-03-01 | 9.37554538E+10 | 0.00025335 |
| price | :float64 | 123 | 0 | 53.01 | 91.26 | | 130.3 | 1.65133647E+01 | 0.44446266 |
| symbol | :string | 123 | 0 | | | IBM | | | |
AMZN: descriptive-stats [3 10]:
| :col-name | :datatype | :n-valid | :n-missing | :min | :mean | :mode | :max | :standard-deviation | :skew |
|-----------|--------------------|----------|------------|------------|------------|-------|------------|---------------------|------------|
| date | :packed-local-date | 123 | 0 | 2000-01-01 | 2005-01-30 | | 2010-03-01 | 9.37554538E+10 | 0.00025335 |
| price | :float64 | 123 | 0 | 5.970 | 47.99 | | 135.9 | 2.88913206E+01 | 0.98217538 |
| symbol | :string | 123 | 0 | | | AMZN | | | |
)
</code></pre>
<h2>Elementwise Operations</h2>
<p>The datatype system includes a mathematical abstraction that is designed to work with
things like columns. Using this we can create new columns that are lazily evaluated
linear combinations of other columns.</p>
<pre><code class="language-clojure">user&gt; (def updated-ames
(assoc ames-ds
"TotalBath"
(dfn/+ (ames-ds "BsmtFullBath")
(dfn/* 0.5 (ames-ds "BsmtHalfBath"))
(ames-ds "FullBath")
(dfn/* 0.5 (ames-ds "HalfBath")))))
#'user/updated-ames
user&gt; (ds/head (ds/select-columns updated-ames ["BsmtFullBath" "BsmtHalfBath" "FullBath" "HalfBath" "TotalBath"]))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 5]:
| BsmtFullBath | BsmtHalfBath | FullBath | HalfBath | TotalBath |
|--------------|--------------|----------|----------|-----------|
| 1 | 0 | 2 | 1 | 3.5 |
| 0 | 1 | 2 | 0 | 2.5 |
| 1 | 0 | 2 | 1 | 3.5 |
| 1 | 0 | 1 | 0 | 2.0 |
| 1 | 0 | 2 | 1 | 3.5 |
</code></pre>
<p>The datatype library contains a typed elementwise-map function named 'emap'
that allows us to do define arbitrary conversions from columns into columns:</p>
<pre><code class="language-clojure">user&gt; (def named-baths (assoc updated-ames "NamedBath" (dtype/emap #(let [tbaths (double %)]
(cond
(&lt; tbaths 1.0)
"almost none"
(&lt; tbaths 2.0)
"somewhat doable"
(&lt; tbaths 3.0)
"getting somewhere"
:else
"living in style"))
:string
(updated-ames "TotalBath"))))
#'user/named-baths
user&gt; (ds/head (ds/select-columns named-baths ["TotalBath" "NamedBath"]))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 2]:
| TotalBath | NamedBath |
|-----------|-------------------|
| 3.5 | living in style |
| 2.5 | getting somewhere |
| 3.5 | living in style |
| 2.0 | getting somewhere |
| 3.5 | living in style |
;; Here we see that the higher level houses all have more bathrooms
user&gt; (def sorted-named-baths (-&gt; (ds/select-columns named-baths ["TotalBath" "NamedBath" "SalePrice"])
(ds/sort-by-column "SalePrice" &gt;)))
#'user/sorted-named-baths
user&gt; (ds/head sorted-named-baths)
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 3]:
| TotalBath | NamedBath | SalePrice |
|-----------|-----------------|-----------|
| 4.0 | living in style | 755000 |
| 4.5 | living in style | 745000 |
| 4.5 | living in style | 625000 |
| 3.5 | living in style | 611657 |
| 3.5 | living in style | 582933 |
user&gt; (ds/tail sorted-named-baths)
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/ames-train.csv.gz [5 3]:
| TotalBath | NamedBath | SalePrice |
|-----------|-----------------|-----------|
| 1.0 | somewhat doable | 40000 |
| 1.0 | somewhat doable | 39300 |
| 1.0 | somewhat doable | 37900 |
| 1.0 | somewhat doable | 35311 |
| 1.0 | somewhat doable | 34900 |
</code></pre>
<h2>DateTime Types</h2>
<p>Support for reading datetime types and manipulating them. Please checkout the
<code>dtype-next</code> <a href="https://cnuernber.github.io/dtype-next/tech.v3.datatype.datetime.html">datetime documentation</a>
right now before moving on :-).</p>
<pre><code class="language-clojure">user&gt; (def stocks (ds/-&gt;dataset "https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv"
{:key-fn keyword}))
#'user/stocks
user&gt; (ds/head stocks)
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv [5 3]:
| :symbol | :date | :price |
|---------|------------|--------|
| MSFT | 2000-01-01 | 39.81 |
| MSFT | 2000-02-01 | 36.35 |
| MSFT | 2000-03-01 | 43.22 |
| MSFT | 2000-04-01 | 28.37 |
| MSFT | 2000-05-01 | 25.45 |
user&gt; (meta (stocks :date))
{:name :date, :datatype :packed-local-date, :n-elems 560}
user&gt; (require '[tech.v3.datatype.datetime :as dtype-dt])
nil
user&gt; (ds/head (ds/update-column stocks :date dtype-dt/datetime-&gt;milliseconds))
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv [5 3]:
| :symbol | :date | :price |
|---------|--------------|--------|
| MSFT | 946684800000 | 39.81 |
| MSFT | 949363200000 | 36.35 |
| MSFT | 951868800000 | 43.22 |
| MSFT | 954547200000 | 28.37 |
| MSFT | 957139200000 | 25.45 |
user&gt; (require '[tech.v3.datatype.functional :as dfn])
nil
user&gt; (as-&gt; (assoc stocks :years (dtype-dt/long-temporal-field :years (stocks :date))) stocks
(ds/group-by stocks (juxt :symbol :years))
(vals stocks)
;;stream is a sequence of datasets at this point.
(map (fn [ds]
{:symbol (first (ds :symbol))
:years (first (ds :years))
:avg-price (dfn/mean (ds :price))})
stocks)
(sort-by (juxt :symbol :years) stocks)
(ds/-&gt;&gt;dataset stocks)
(ds/head stocks 10))
_unnamed [10 3]:
| :symbol | :years | :avg-price |
|---------|--------|--------------|
| AAPL | 2000 | 21.74833333 |
| AAPL | 2001 | 10.17583333 |
| AAPL | 2002 | 9.40833333 |
| AAPL | 2003 | 9.34750000 |
| AAPL | 2004 | 18.72333333 |
| AAPL | 2005 | 48.17166667 |
| AAPL | 2006 | 72.04333333 |
| AAPL | 2007 | 133.35333333 |
| AAPL | 2008 | 138.48083333 |
| AAPL | 2009 | 150.39333333 |
</code></pre>
<h2>Writing A Dataset Out</h2>
<p>These forms are supported for writing out a dataset:</p>
<pre><code class="language-clojure">(ds/write! test-ds "test.csv")
(ds/write! test-ds "test.tsv")
(ds/write! test-ds "test.tsv.gz")
(ds/write! test-ds "test.nippy")
(ds/write! test-ds out-stream)
</code></pre>
<p>We have good support for <a href="nippy-serialization-rocks.html">nippy</a> in which case
datasets work just like any other datastructure. This format allows some level of
compression but about 10X-100X the loading performance of gzipped csv/tsv. In addition,
you can write out heterogeneous datastructures that contain datasets and other things
such as the result of a group-by:</p>
<pre><code class="language-clojure">user&gt; (require '[taoensso.nippy :as nippy])
nil
user&gt; (def byte-data (nippy/freeze (ds/group-by stocks :symbol)))
#'user/byte-data
user&gt; (type byte-data)
[B
user&gt; (keys (nippy/thaw byte-data))
("MSFT" "GOOG" "AAPL" "IBM" "AMZN")
user&gt; (first (nippy/thaw byte-data))
["MSFT"
https://github.com/techascent/tech.ml.dataset/raw/master/test/data/stocks.csv [123 3]:
| :symbol | :date | :price |
|---------|------------|--------|
| MSFT | 2000-01-01 | 39.81 |
| MSFT | 2000-02-01 | 36.35 |
| MSFT | 2000-03-01 | 43.22 |
| MSFT | 2000-04-01 | 28.37 |
| MSFT | 2000-05-01 | 25.45 |
</code></pre>
<p>Also see the <a href="https://techascent.github.io/tech.ml.dataset/tech.v3.libs.arrow.html">tech.v3.libs.arrow</a> and
<a href="https://techascent.github.io/tech.ml.dataset/tech.v3.libs.parquet.html">tech.v3.libs.parquet</a> namespaces.</p>
<p>If you made it this far, check out the <a href="https://techascent.github.io/tech.ml.dataset/quick-reference.html">quick reference</a>.</p>
</div></div></div></body></html>
Reference in New Issue View Git Blame Copy Permalink