methods/svec/src/pg_gp/svec.sql_in

/* ----------------------------------------------------------------------- *//**
 *
 * @file svec.sql_in
 *
 * @brief SQL type definitions and functions for sparse vector data type
 *        <tt>svec</tt>
 *
 * @sa For an introduction to the sparse vector implementation, see the module
 *     description \ref grp_svec.
 *
 *//* ----------------------------------------------------------------------- */

m4_include(`SQLCommon.m4')

/**
@addtogroup grp_svec

<div class="toc"><b>Contents</b>
<ul>
<li><a href="#usage">Using Sparse Vectors</a></li>
<li><a href="#vectorization">Document Vectorization into Sparse Vectors</a></li>
<li><a href="#examples">Examples</a></li>
<li><a href="#related">Related Topics</a></li>
</ul>
</div>

@brief Implements a sparse vector data type that provides compressed storage
of vectors that may have many duplicate elements.

This module implements a sparse vector data type, named "svec", which
provides compressed storage of vectors that have many duplicate elements.

Arrays of floating point numbers for various calculations  sometimes have
long runs of zeros (or some other default value).  This is common in
applications like scientific computing,  retail optimization, and text
processing. Each floating point number takes  8 bytes of storage in memory
and/or disk, so saving those zeros is often  worthwhile. There are also
many computations that can benefit from skipping over the zeros.

Consider, for example, the following
array of doubles stored as a Postgres/Greenplum "float8[]" data type:

<pre class="example">
'{0, 33,...40,000 zeros..., 12, 22 }'\::float8[]
</pre>

This array would occupy slightly more than 320KB of memory or disk, most of
it zeros. Even if we were to exploit the null bitmap and store the zeros
as nulls, we would still end up with a 5KB null bitmap, which is still not
nearly as memory efficient as we'd like. Also, as we perform various
operations on the array, we do work on 40,000 fields that turn out to be
unimportant.

To solve the problems associated with the processing of vectors discussed
above, the svec type employs a simple Run Length Encoding (RLE) scheme to
represent sparse vectors as pairs of count-value arrays. For example, the
array above would be represented as

<pre class="example">
'{1,1,40000,1,1}:{0,33,0,12,22}'\::madlib.svec
</pre>

which says there is 1 occurrence of 0, followed by 1 occurrence of 33,
followed by 40,000 occurrences of 0, etc. This uses just 5 integers and 5
floating point numbers to store the array. Further, it is easy to
implement vector operations that can take advantage of the RLE
representation to make computations faster. The SVEC module provides a library
of such functions.

The current version only supports sparse vectors of float8 values. Future
versions will support other base types.

@anchor usage
@par Using Sparse Vectors

An SVEC can be constructed directly with a constant expression, as follows:
<pre class="example">
SELECT '{n1,n2,...,nk}:{v1,v2,...vk}'\::madlib.svec;
</pre>
where <tt>n1,n2,...,nk</tt> specifies the counts for the values <tt>v1,v2,...,vk</tt>.

A float array can be cast to an SVEC:
<pre class="example">
SELECT ('{v1,v2,...vk}'\::float[])\::madlib.svec;
</pre>

An SVEC can be created with an aggregation:
<pre class="example">
SELECT madlib.svec_agg(v1) FROM generate_series(1,k);
</pre>

An SVEC can be created using the
<tt>madlib.svec_cast_positions_float8arr()</tt> function by supplying an
array of positions and an array of values at those positions:
<pre class="example">
SELECT madlib.svec_cast_positions_float8arr(
    array[n1,n2,...nk],    -- positions of values in vector
    array[v1,v2,...vk],    -- values at each position
    length,                -- length of vector
    base)                  -- value at unspecified positions
</pre>
For example, the following expression:
<pre class="example">
SELECT madlib.svec_cast_positions_float8arr(
    array[1,3,5],
    array[2,4,6],
    10,
    0.0)
</pre>
produces this SVEC:
<pre class="result">
 svec_cast_positions_float8arr
 &nbsp;------------------------------
 {1,1,1,1,1,5}:{2,0,4,0,6,0}
</pre>

Add MADLIB_SCHEMA to the search_path to use the svec operators
defined in the module.

@anchor vectorization
@par Document Vectorization into Sparse Vectors
This module implements an efficient way for document vectorization, converting
text documents into sparse vector representation (MADlib.svec), required by various
machine learning algorithms in MADlib.

The function accepts two tables as input, dictionary table and documents
table, and produces the specified output table containing sparse vectors
for the represented documents (in documents table).

<pre class="syntax">
madlib.gen_doc_svecs(output_tbl,
                     dictionary_tbl,
                     dict_id_col,
                     dict_term_col,
                     documents_tbl,
                     doc_id_col,
                     doc_term_col,
                     doc_term_info_col
                    )
</pre>
\b Arguments
<DL class="arglist">
<DT>output_tbl</DT>
<DD>TEXT. Name of the output table to be created containing the sparse
vector representation of the documents. It has the following columns:
    <table class="output">
    <tr>
        <th>doc_id</th>
        <td>__TYPE_DOC__. Document id.
            \n \__TYPE_DOC__: Column type depends on the type of \c doc_id_col in \c documents_tbl.
        </td>
    </tr>
    <tr>
        <th>sparse_vector</th>
        <td>MADlib.svec. Corresponding sparse vector representation.</td>
    </tr>
    </table>
</DD>

<DT>dictionary_tbl</DT>
<DD>TEXT. Name of the dictionary table containing features.
    <table class="output">
    <tr>
        <th>dict_id_col</th>
        <td>TEXT. Name of the id column in the \c dictionary_tbl.
            \n Expected Type: INTEGER or BIGINT.
            \n NOTE: Values must be continuous ranging from 0 to total number of elements in
            the dictionary - 1.</dt>
    </tr>
    <tr>
        <th>dict_term_col</th>
        <td>TEXT. Name of the column containing term (features) in \c dictionary_tbl.</td>
    </tr>
    </table>
</DD>

<DT>documents_tbl</DT>
<DD>TEXT. Name of the documents table representing documents.
    <table class="output">
    <tr>
        <th>doc_id_col</th>
        <td>TEXT. Name of the id column in the \c documents_tbl.</td>
    </tr>
    <tr>
        <th>doc_term_col</th>
        <td>TEXT. Name of the term column in the \c documents_tbl.</td>
    </tr>
    <tr>
        <th>doc_term_info_col</th>
        <td>TEXT. Name of the term info column in \c documents_tbl. The expected type of this column should be:
            \n - INTEGER, BIGINT or DOUBLE PRECISION: Values directly used to populate vector.
            \n - ARRAY: Length of the array used to populate the vector.
            \n ** For an example use case on using these types of column types, please refer to the example below.
        </td>
    </tr>
    </table>
</DD>
</DL>

\b Example:
\n Consider a corpus consisting of set of documents consisting of features (terms) along with doc ids:
<pre class="example">
1, {this,is,one,document,in,the,corpus}
2, {i,am,the,second,document,in,the,corpus}
3, {being,third,never,really,bothered,me,until,now}
4, {the,document,before,me,is,the,third,document}
</pre>

-# Prepare documents table in appropriate format.
\n The corpus specified above can be represented by any of the following \c documents_table:
<pre class="example">
SELECT * FROM documents_table ORDER BY id;
</pre>
Result:
<pre class="result">
  id |   term   | count                 id |   term   | positions
&nbsp;----+----------+-------               ----+----------+-----------
   1 | is       |     1                  1 | is       | {1}
   1 | in       |     1                  1 | in       | {4}
   1 | one      |     1                  1 | one      | {2}
   1 | this     |     1                  1 | this     | {0}
   1 | the      |     1                  1 | the      | {5}
   1 | document |     1                  1 | document | {3}
   1 | corpus   |     1                  1 | corpus   | {6}
   2 | second   |     1                  2 | second   | {3}
   2 | document |     1                  2 | document | {4}
   2 | corpus   |     1                  2 | corpus   | {7}
   . | ...      |    ..                  . | ...      | ...
   4 | document |     2                  4 | document | {1,7}
...
</pre>

-# Prepare dictionary table in appropriate format.
<pre class="example">
SELECT * FROM dictionary_table ORDER BY id;
</pre>
Result:
<pre class="result">
  id |   term
&nbsp;----+----------
   0 | am
   1 | before
   2 | being
   3 | bothered
   4 | corpus
   5 | document
   6 | i
   7 | in
   8 | is
   9 | me
...
</pre>

-# Generate sparse vector for the documents using dictionary_table and documents_table.
\n \c doc_term_info_col \c (count) of type INTEGER:
<pre class="example">
SELECT * FROM madlib.gen_doc_svecs('svec_output', 'dictionary_table', 'id', 'term',
                            'documents_table', 'id', 'term', 'count');
</pre>
\c  doc_term_info_col \c (positions) of type ARRAY:
<pre class="example">
SELECT * FROM madlib.gen_doc_svecs('svec_output', 'dictionary_table', 'id', 'term',
                            'documents_table', 'id', 'term', 'positions');
</pre>
Result:
<pre class="result">
                                 gen_doc_svecs
&nbsp;--------------------------------------------------------------------------------------
 Created table svec_output (doc_id, sparse_vector) containing sparse vectors
(1 row)
</pre>

-# Analyze the sparse vectors created.
<pre class="example">
SELECT * FROM svec_output ORDER by doc_id;
</pre>
Result:
<pre class="result">
 doc_id |                  sparse_vector
&nbsp;--------+-------------------------------------------------
      1 | {4,2,1,2,3,1,2,1,1,1,1}:{0,1,0,1,0,1,0,1,0,1,0}
      2 | {1,3,4,6,1,1,3}:{1,0,1,0,1,2,0}
      3 | {2,2,5,3,1,1,2,1,1,1}:{0,1,0,1,0,1,0,1,0,1}
      4 | {1,1,3,1,2,2,5,1,1,2}:{0,1,0,2,0,1,0,2,1,0}
(4 rows)
</pre>

See the file svec.sql_in for complete syntax.

@anchor examples
@examp

We can use operations with svec type like <, >, *, **, /, =, +, SUM, etc,
and they have meanings associated with typical vector operations. For
example, the plus (+) operator adds each of the terms of two vectors having
the same dimension together.
<pre class="example">
SELECT ('{0,1,5}'\::float8[]\::madlib.svec + '{4,3,2}'\::float8[]\::madlib.svec)\::float8[];
</pre>
Result:
<pre class="result">
 float8
&nbsp;--------
 {4,4,7}
</pre>

Without the casting into float8[] at the end, we get:
<pre class="example">
SELECT '{0,1,5}'\::float8[]\::madlib.svec + '{4,3,2}'\::float8[]\::madlib.svec;
</pre>
Result:
<pre class="result">
 ?column?
&nbsp;---------
{2,1}:{4,7}
</pre>

A dot product (%*%) between the two vectors will result in a scalar
result of type float8. The dot product should be (0*4 + 1*3 + 5*2) = 13,
like this:
<pre class="example">
SELECT '{0,1,5}'\::float8[]\::madlib.svec %*% '{4,3,2}'\::float8[]\::madlib.svec;
</pre>
<pre class="result">
 ?column?
&nbsp;---------
	13
</pre>

Special vector aggregate functions are also available. SUM is self
explanatory. SVEC_COUNT_NONZERO evaluates the count of non-zero terms
in each column found in a set of n-dimensional svecs and returns an
svec with the counts. For instance, if we have the vectors {0,1,5},
{10,0,3},{0,0,3},{0,1,0}, then executing the SVEC_COUNT_NONZERO() aggregate
function would result in {1,2,3}:

<pre class="example">
CREATE TABLE list (a madlib.svec);
INSERT INTO list VALUES ('{0,1,5}'\::float8[]), ('{10,0,3}'\::float8[]), ('{0,0,3}'\::float8[]),('{0,1,0}'\::float8[]);
SELECT madlib.svec_count_nonzero(a)\::float8[] FROM list;
</pre>
Result:
<pre class="result">
svec_count_nonzero
&nbsp;----------------
	{1,2,3}
</pre>

We do not use null bitmaps in the svec data type. A null value in an svec
is represented explicitly as an NVP (No Value Present) value. For example,
we have:
<pre class="example">
SELECT '{1,2,3}:{4,null,5}'\::madlib.svec;
</pre>
Result:
<pre class="result">
	  svec
&nbsp;------------------
 {1,2,3}:{4,NVP,5}
</pre>

Adding svecs with null values results in NVPs in the sum:
<pre class="example">
SELECT '{1,2,3}:{4,null,5}'\::madlib.svec + '{2,2,2}:{8,9,10}'\::madlib.svec;
</pre>
Result:
<pre class="result">
		 ?column?
 &nbsp;-------------------------
  {1,2,1,2}:{12,NVP,14,15}
</pre>

An element of an svec can be accessed using the svec_proj() function,
which takes an svec and the index of the element desired.
<pre class="example">
SELECT madlib.svec_proj('{1,2,3}:{4,5,6}'\::madlib.svec, 1) + madlib.svec_proj('{4,5,6}:{1,2,3}'\::madlib.svec, 15);
</pre>
Result:
<pre class="result"> ?column?
&nbsp;---------
	7
</pre>

A subvector of an svec can be accessed using the svec_subvec() function,
which takes an svec and the start and end index of the subvector desired.
<pre class="example">
SELECT madlib.svec_subvec('{2,4,6}:{1,3,5}'\::madlib.svec, 2, 11);
</pre>
Result:
<pre class="result">   svec_subvec
&nbsp;----------------
 {1,4,5}:{1,3,5}
</pre>

The elements/subvector of an svec can be changed using the function
svec_change(). It takes three arguments: an m-dimensional svec sv1, a
start index j, and an n-dimensional svec sv2 such that j + n - 1 <= m,
and returns an svec like sv1 but with the subvector sv1[j:j+n-1]
replaced by sv2. An example follows:
<pre class="example">
SELECT madlib.svec_change('{1,2,3}:{4,5,6}'\::madlib.svec,3,'{2}:{3}'\::madlib.svec);
</pre>
Result:
<pre class="result">	 svec_change
&nbsp;--------------------
 {1,1,2,2}:{4,5,3,6}
</pre>

There are also higher-order functions for processing svecs. For example,
the following is the corresponding function for lapply() in R.
<pre class="example">
SELECT madlib.svec_lapply('sqrt', '{1,2,3}:{4,5,6}'\::madlib.svec);
</pre>
Result:
<pre class="result">
				  svec_lapply
&nbsp;----------------------------------------------
 {1,2,3}:{2,2.23606797749979,2.44948974278318}
</pre>

The full list of functions available for operating on svecs are available
in svec.sql-in.

<b> A More Extensive Example</b>

For a text classification example, let's assume we have a dictionary
composed of words in a sorted text array:
<pre class="example">
CREATE TABLE features (a text[]);
INSERT INTO features VALUES
			('{am,before,being,bothered,corpus,document,i,in,is,me,
			   never,now,one,really,second,the,third,this,until}');
</pre>
    We have a set of documents, each represented as an array of words:
<pre class="example">
CREATE TABLE documents(a int,b text[]);
INSERT INTO documents VALUES
			(1,'{this,is,one,document,in,the,corpus}'),
			(2,'{i,am,the,second,document,in,the,corpus}'),
			(3,'{being,third,never,really,bothered,me,until,now}'),
			(4,'{the,document,before,me,is,the,third,document}');
</pre>

Now we have a dictionary and some documents, we would like to do some
document categorization using vector arithmetic on word counts and
proportions of dictionary words in each document.

To start this process, we'll need to find the dictionary words in each
document. We'll prepare what is called a Sparse Feature Vector or SFV
for each document. An SFV is a vector of dimension N, where N is the
number of dictionary words, and in each cell of an SFV is a count of
each dictionary word in the document.

Inside the sparse vector library, we have a function that will create
an SFV from a document, so we can just do this (For a more efficient
way for converting documents into sparse vectors, especially for larger
datasets, please refer to <li><a href="#vectorization">Document
Vectorization into Sparse Vectors</a></li>):

<pre class="example">
SELECT madlib.svec_sfv((SELECT a FROM features LIMIT 1),b)\::float8[]
		 FROM documents;
</pre>
Result:
<pre class="result">
				svec_sfv
&nbsp;----------------------------------------
 {0,0,0,0,1,1,0,1,1,0,0,0,1,0,0,1,0,1,0}
 {0,0,1,1,0,0,0,0,0,1,1,1,0,1,0,0,1,0,1}
 {1,0,0,0,1,1,1,1,0,0,0,0,0,0,1,2,0,0,0}
 {0,1,0,0,0,2,0,0,1,1,0,0,0,0,0,2,1,0,0}
</pre>

Note that the output of madlib.svec_sfv() is an svec for each
document containing the count of each of the dictionary words in the
ordinal positions of the dictionary. This can more easily be understood
by lining up the feature vector and text like this:

<pre class="example">
SELECT madlib.svec_sfv((SELECT a FROM features LIMIT 1),b)\::float8[]
				, b
		 FROM documents;
</pre>
Result:
<pre class="result">
				svec_sfv                 |                        b
&nbsp;----------------------------------------+--------------------------------------------------
 {1,0,0,0,1,1,1,1,0,0,0,0,0,0,1,2,0,0,0} | {i,am,the,second,document,in,the,corpus}
 {0,1,0,0,0,2,0,0,1,1,0,0,0,0,0,2,1,0,0} | {the,document,before,me,is,the,third,document}
 {0,0,0,0,1,1,0,1,1,0,0,0,1,0,0,1,0,1,0} | {this,is,one,document,in,the,corpus}
 {0,0,1,1,0,0,0,0,0,1,1,1,0,1,0,0,1,0,1} | {being,third,never,really,bothered,me,until,now}
</pre>
<pre class="example">
SELECT * FROM features;
</pre>
<pre class="result">
												a
&nbsp;-------------------------------------------------------------------------------------------------------
{am,before,being,bothered,corpus,document,i,in,is,me,never,now,one,really,second,the,third,this,until}
</pre>

Now when we look at the document "i am the second document in the corpus",
its SFV is {1,3*0,1,1,1,1,6*0,1,2}. The word "am" is the first ordinate in
the dictionary and there is 1 instance of it in the SFV. The word "before"
has no instances in the document, so its value is "0" and so on.

The function madlib.svec_sfv() can process large
numbers of documents into their SFVs in parallel at high speed.

The rest of the categorization process is all vector math. The actual
count is hardly ever used.  Instead, it's turned into a weight. The most
common weight is called tf/idf for Term Frequency / Inverse Document
Frequency. The calculation for a given term in a given document is

<pre class="example">
{\#Times in document} * log {\#Documents / \#Documents the term appears in}.
</pre>

For instance, the term "document" in document A would have weight
1 * log (4/3). In document D, it would have weight 2 * log (4/3).
Terms that appear in every document would have tf/idf weight 0, since
log (4/4) = log(1) = 0. (Our example has no term like that.) That
usually sends a lot of values to 0.

For this part of the processing, we'll need to have a sparse vector of
the dictionary dimension (19) with the values
<pre class="example">
log(\#documents/\#Documents each term appears in).
</pre>
There will be one such vector for the whole list of documents (aka the
"corpus"). The \#documents is just a count of all of the documents, in
this case 4, but there is one divisor for each dictionary word and its
value is the count of all the times that word appears in the document.
This single vector for the whole corpus can then be scalar product
multiplied by each document SFV to produce the Term Frequency/Inverse
Document Frequency weights.

This can be done as follows:
<pre class="example">
CREATE TABLE corpus AS
			(SELECT a, madlib.svec_sfv((SELECT a FROM features LIMIT 1),b) sfv
		 FROM documents);
CREATE TABLE weights AS
		  (SELECT a docnum, madlib.svec_mult(sfv, logidf) tf_idf
		   FROM (SELECT madlib.svec_log(madlib.svec_div(count(sfv)\::madlib.svec,madlib.svec_count_nonzero(sfv))) logidf
				FROM corpus) foo, corpus ORDER BYdocnum);
SELECT * FROM weights;
</pre>
Result
<pre class="result">
docnum |                tf_idf
&nbsp;------+----------------------------------------------------------------------
	 1 | {4,1,1,1,2,3,1,2,1,1,1,1}:{0,0.69,0.28,0,0.69,0,1.38,0,0.28,0,1.38,0}
	 2 | {1,3,1,1,1,1,6,1,1,3}:{1.38,0,0.69,0.28,1.38,0.69,0,1.38,0.57,0}
	 3 | {2,2,5,1,2,1,1,2,1,1,1}:{0,1.38,0,0.69,1.38,0,1.38,0,0.69,0,1.38}
	 4 | {1,1,3,1,2,2,5,1,1,2}:{0,1.38,0,0.57,0,0.69,0,0.57,0.69,0}
</pre>

We can now get the "angular distance" between one document and the rest
of the documents using the ACOS of the dot product of the document vectors:
The following calculates the angular distance between the first document
and each of the other documents:
<pre class="example">
SELECT docnum,
				180. * ( ACOS( madlib.svec_dmin( 1., madlib.svec_dot(tf_idf, testdoc)
					/ (madlib.svec_l2norm(tf_idf)*madlib.svec_l2norm(testdoc))))/3.141592654) angular_distance
		 FROM weights,(SELECT tf_idf testdoc FROM weights WHERE docnum = 1 LIMIT 1) foo
		 ORDER BY 1;
</pre>
Result:
<pre class="result">
docnum | angular_distance
&nbsp;-------+------------------
	 1 |                0
	 2 | 78.8235846096986
	 3 | 89.9999999882484
	 4 | 80.0232034288617
</pre>

We can see that the angular distance between document 1 and itself
is 0 degrees and between document 1 and 3 is 90 degrees because they
share no features at all. The angular distance can now be plugged into
machine learning algorithms that rely on a distance measure between
data points.

SVEC also provides functionality for declaring array given an array of
positions and array of values, intermediate values betweens those are
declared to be base value that user provides in the same function call. In
the example below the fist array of integers represents the positions for
the array two (array of floats). Positions do not need to come in the
sorted order.  Third value represents desired maximum size of the array.
This assures that array is of that size even if last position is not. If
max size < 1 that value is ignored and array will end at the last position
in the position vector. Final value is a float representing the
base value to be used between the declared ones (0 would be a common
candidate):

<pre class="example">
SELECT madlib.svec_cast_positions_float8arr(ARRAY[1,2,7,5,87],ARRAY[.1,.2,.7,.5,.87],90,0.0);
</pre>
Result:
<pre class="result">
		svec_cast_positions_float8arr
&nbsp;----------------------------------------------------
{1,1,2,1,1,1,79,1,3}:{0.1,0.2,0,0.5,0,0.7,0,0.87,0}
(1 row)
</pre>


@anchor related
@par Related Topics

Other examples of svecs usage can be found in the k-means module, \ref grp_kmeans.

File svec.sql_in documenting the SQL functions.

@internal
@sa File sparse_vector.c documenting the implementation in C.
@endinternal
*/


--! @file svec.sql_in
--!

-- DROP SCHEMA MADLIB_SCHEMA CASCADE;
-- CREATE SCHEMA MADLIB_SCHEMA;

-- DROP TYPE IF EXISTS MADLIB_SCHEMA.svec CASCADE;
CREATE TYPE MADLIB_SCHEMA.svec;

--! SVEC constructor from CSTRING.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_in(cstring)
    RETURNS MADLIB_SCHEMA.svec
    AS 'MODULE_PATHNAME'
    LANGUAGE C IMMUTABLE STRICT
    m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Converts SVEC to CSTRING.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_out(MADLIB_SCHEMA.svec)
    RETURNS cstring
    AS 'MODULE_PATHNAME'
    LANGUAGE C IMMUTABLE STRICT
    m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Converts SVEC internal representation to SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_recv(internal)
    RETURNS MADLIB_SCHEMA.svec
    AS 'MODULE_PATHNAME'
    LANGUAGE C IMMUTABLE STRICT
    m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Converts SVEC to BYTEA.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_send(MADLIB_SCHEMA.svec)
    RETURNS bytea
    AS 'MODULE_PATHNAME'
    LANGUAGE C IMMUTABLE STRICT
    m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

CREATE TYPE MADLIB_SCHEMA.svec (
       internallength = VARIABLE,
       input = MADLIB_SCHEMA.svec_in,
       output = MADLIB_SCHEMA.svec_out,
       send = MADLIB_SCHEMA.svec_send,
       receive = MADLIB_SCHEMA.svec_recv,
       storage=EXTENDED,
       alignment = double
);

--! Adds two SVECs together, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_plus(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_plus' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Minus second SVEC from the first, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_minus(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_minus' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Divides the first SVEC by the second, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_div(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_div' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Multiplies two SVEVs together, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_mult(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_mult' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Raises each element of the first SVEC to the power given by second SVEC, which must have dimension 1 (a scalar).
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_pow(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_pow' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Compare two SVECs, -1 for left side smaller, 1 for greater, 0 for equal
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cmp(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS integer AS 'MODULE_PATHNAME', 'svec_cmp' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the left SVEC is greater or equal to the right.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_ge(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_ge' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the left SVEC is less than the right.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_lt(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_lt' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the left SVEC is less or equal to the right.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_le(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_le' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the left SVEC is greater.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_gt(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_gt' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if two SVECs are not equal. If the two SVECs are of different size, then will return true.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_ne(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_ne' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if two SVECs are equal. If the two SVECs are of different size, then will return false.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_eq(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS boolean AS 'MODULE_PATHNAME', 'svec_eq' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if two float8 arrays are equal
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_eq(float8[],float8[]) RETURNS boolean AS 'MODULE_PATHNAME', 'float8arr_equals' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Minus second array from the first array, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_minus_float8arr(float8[],float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_minus_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Minus second SVEC from the first array, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_minus_svec(float8[],MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_minus_svec' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Minus second array from the first SVEC, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_minus_float8arr(MADLIB_SCHEMA.svec,float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_minus_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Adds two arrays together, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_plus_float8arr(float8[],float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_plus_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Adds an array and an SVEC, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_plus_svec(float8[],MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_plus_svec' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Adds an SVEC and an array, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_plus_float8arr(MADLIB_SCHEMA.svec,float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_plus_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Multiplies two float8 arrays, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_mult_float8arr(float8[],float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_mult_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Multiplies an array and an SVEC, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_mult_svec(float8[],MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_mult_svec' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Multiplies an SVEC and an array, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_mult_float8arr(MADLIB_SCHEMA.svec,float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_mult_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Divides a float8 array by another, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_div_float8arr(float8[],float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_div_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Divides a float8 array by an SVEC, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_div_svec(float8[],MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'float8arr_div_svec' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Divides an SVEC by a float8 array, element by element.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_div_float8arr(MADLIB_SCHEMA.svec,float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_div_float8arr' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Computes the dot product of two SVECs.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_dot(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS float8 AS 'MODULE_PATHNAME', 'svec_dot' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Computes the dot product of two float8 arrays.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_dot(float8[],float8[]) RETURNS float8 AS 'MODULE_PATHNAME', 'float8arr_dot' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Computes the dot product of an SVEC and a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_dot(MADLIB_SCHEMA.svec,float8[]) RETURNS float8 AS 'MODULE_PATHNAME', 'svec_dot_float8arr' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Computes the dot product of a float8 array and an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_dot(float8[],MADLIB_SCHEMA.svec) RETURNS float8 AS 'MODULE_PATHNAME', 'float8arr_dot_svec' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int2 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_int2(int2) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_int2' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int4 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_int4(int4) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_int4' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int8 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_int8(bigint) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_int8' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a float4 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_float4(float4) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_float4' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a float8 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_float8(float8) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_float8' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a numeric into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_numeric(numeric) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_numeric' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int2 into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_int2(int2) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_int2' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int4 into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_int4(int4) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_int4' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an int8 into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_int8(bigint) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_int8' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a float4 into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_float4(float4) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_float4' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a float8 into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_float8(float8) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_float8' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a numeric into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.float8arr_cast_numeric(numeric) RETURNS float8[] AS 'MODULE_PATHNAME', 'float8arr_cast_numeric' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts a float8 into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_float8arr(float8[]) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_float8arr' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an array of int8 positions, float8 values into an SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_cast_positions_float8arr(int8[],float8[],int8,float8) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_cast_positions_float8arr' STRICT LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Casts an SVEC into a float8 array.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_return_array(MADLIB_SCHEMA.svec) RETURNS float8[] AS 'MODULE_PATHNAME', 'svec_return_array' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Concatenates two SVECs.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_concat(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_concat' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Replicates n copies of an SVEC and concatenates them together.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_concat_replicate(int4,MADLIB_SCHEMA.svec) RETURNS MADLIB_SCHEMA.svec AS 'MODULE_PATHNAME', 'svec_concat_replicate' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

/*
DROP OPERATOR IF EXISTS || ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS - ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS + ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS / ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS %*% ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS * ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS ^ ( MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);
*/

CREATE OPERATOR MADLIB_SCHEMA.|| (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_concat
);

CREATE OPERATOR MADLIB_SCHEMA.- (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_minus
);
CREATE OPERATOR MADLIB_SCHEMA.+ (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_plus
);
CREATE OPERATOR MADLIB_SCHEMA./ (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_div
);
CREATE OPERATOR MADLIB_SCHEMA.%*% (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_dot
);
CREATE OPERATOR MADLIB_SCHEMA.* (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_mult
);
CREATE OPERATOR MADLIB_SCHEMA.^ (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_pow
);

-- float8[] operators
-- DROP OPERATOR IF EXISTS = ( float8[], float8[]);
/*
DROP OPERATOR IF EXISTS %*% ( float8[], MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS %*% ( MADLIB_SCHEMA.svec, float8[]);
DROP OPERATOR IF EXISTS %*% ( float8[], float8[]);
DROP OPERATOR IF EXISTS - ( float8[], float8[]);
DROP OPERATOR IF EXISTS + ( float8[], float8[]);
DROP OPERATOR IF EXISTS * ( float8[], float8[]);
DROP OPERATOR IF EXISTS / ( float8[], float8[]);
DROP OPERATOR IF EXISTS - ( float8[], MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS + ( float8[], MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS * ( float8[], MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS / ( float8[], MADLIB_SCHEMA.svec);
DROP OPERATOR IF EXISTS - ( MADLIB_SCHEMA.svec, float8[]);
DROP OPERATOR IF EXISTS + ( MADLIB_SCHEMA.svec, float8[]);
DROP OPERATOR IF EXISTS * ( MADLIB_SCHEMA.svec, float8[]);
DROP OPERATOR IF EXISTS / ( MADLIB_SCHEMA.svec, float8[]);
*/

/*
CREATE OPERATOR MADLIB_SCHEMA.= (
	leftarg = float8[],
	rightarg = float8[],
	procedure = MADLIB_SCHEMA.float8arr_eq,
	commutator = operator(MADLIB_SCHEMA.=) ,
--	negator = operator(MADLIB_SCHEMA.<>) ,
	restrict = eqsel, join = eqjoinsel
);
*/

CREATE OPERATOR MADLIB_SCHEMA.%*% (
	LEFTARG = float8[],
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_dot
);
CREATE OPERATOR MADLIB_SCHEMA.%*% (
	LEFTARG = float8[],
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.svec_dot
);
CREATE OPERATOR MADLIB_SCHEMA.%*% (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_dot
);
CREATE OPERATOR MADLIB_SCHEMA.- (
	LEFTARG = float8[],
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.float8arr_minus_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA.+ (
	LEFTARG = float8[],
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.float8arr_plus_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA.* (
	LEFTARG = float8[],
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.float8arr_mult_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA./ (
	LEFTARG = float8[],
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.float8arr_div_float8arr
);

CREATE OPERATOR MADLIB_SCHEMA.- (
	LEFTARG = float8[],
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.float8arr_minus_svec
);
CREATE OPERATOR MADLIB_SCHEMA.+ (
	LEFTARG = float8[],
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.float8arr_plus_svec
);
CREATE OPERATOR MADLIB_SCHEMA.* (
	LEFTARG = float8[],
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.float8arr_mult_svec
);
CREATE OPERATOR MADLIB_SCHEMA./ (
	LEFTARG = float8[],
	RIGHTARG = MADLIB_SCHEMA.svec,
	PROCEDURE = MADLIB_SCHEMA.float8arr_div_svec
);

CREATE OPERATOR MADLIB_SCHEMA.- (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_minus_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA.+ (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_plus_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA.* (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_mult_float8arr
);
CREATE OPERATOR MADLIB_SCHEMA./ (
	LEFTARG = MADLIB_SCHEMA.svec,
	RIGHTARG = float8[],
	PROCEDURE = MADLIB_SCHEMA.svec_div_float8arr
);

/*
DROP CAST IF EXISTS (int2 AS MADLIB_SCHEMA.svec) ;
DROP CAST IF EXISTS (integer AS MADLIB_SCHEMA.svec) ;
DROP CAST IF EXISTS (bigint AS MADLIB_SCHEMA.svec) ;
DROP CAST IF EXISTS (float4 AS MADLIB_SCHEMA.svec) ;
DROP CAST IF EXISTS (float8 AS MADLIB_SCHEMA.svec) ;
DROP CAST IF EXISTS (numeric AS MADLIB_SCHEMA.svec) ;
*/

CREATE CAST (int2 AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_int2(int2) ; -- AS IMPLICIT;
CREATE CAST (integer AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_int4(integer) ; -- AS IMPLICIT;
CREATE CAST (bigint AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_int8(bigint) ; -- AS IMPLICIT;
CREATE CAST (float4 AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_float4(float4) ; -- AS IMPLICIT;
CREATE CAST (float8 AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_float8(float8) ; -- AS IMPLICIT;
CREATE CAST (numeric AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_numeric(numeric) ; -- AS IMPLICIT;

/*
DROP CAST IF EXISTS (int2 AS float8[]) ;
DROP CAST IF EXISTS (integer AS float8[]) ;
DROP CAST IF EXISTS (bigint AS float8[]) ;
DROP CAST IF EXISTS (float4 AS float8[]) ;
DROP CAST IF EXISTS (float8 AS float8[]) ;
DROP CAST IF EXISTS (numeric AS float8[]) ;
*/

-- CREATE CAST (int2 AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_int2(int2) ; -- AS IMPLICIT;
-- CREATE CAST (integer AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_int4(integer) ; -- AS IMPLICIT;
-- CREATE CAST (bigint AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_int8(bigint) ; -- AS IMPLICIT;
-- CREATE CAST (float4 AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_float4(float4) ; -- AS IMPLICIT;
-- CREATE CAST (float8 AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_float8(float8) ; -- AS IMPLICIT;
-- CREATE CAST (numeric AS float8[]) WITH FUNCTION MADLIB_SCHEMA.float8arr_cast_numeric(numeric) ; -- AS IMPLICIT;

-- DROP CAST IF EXISTS (MADLIB_SCHEMA.svec AS float8[]) ;
-- DROP CAST IF EXISTS (float8[] AS MADLIB_SCHEMA.svec) ;

CREATE CAST (MADLIB_SCHEMA.svec AS float8[]) WITH FUNCTION MADLIB_SCHEMA.svec_return_array(MADLIB_SCHEMA.svec) ; -- AS IMPLICIT;
CREATE CAST (float8[] AS MADLIB_SCHEMA.svec) WITH FUNCTION MADLIB_SCHEMA.svec_cast_float8arr(float8[]) ; -- AS IMPLICIT;

-- DROP OPERATOR IF EXISTS = (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) ;


CREATE OPERATOR MADLIB_SCHEMA.= (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_eq,
	commutator = operator(MADLIB_SCHEMA.=) ,
--	negator = operator(MADLIB_SCHEMA.<>) ,
	restrict = eqsel, join = eqjoinsel
);

-- Comparisons based on L2 Norm
--! Returns true if the l2 norm of the first SVEC is less than that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_lt(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_lt' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the l2 norm of the first SVEC is less than or equal to that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_le(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_le' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the l2 norm of the first SVEC is equal to that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_eq(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_eq' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the l2 norm of the first SVEC is not equal to that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_ne(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_ne' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the l2 norm of the first SVEC is greater than that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_gt(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_gt' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns true if the l2 norm of the first SVEC is greater than or equal to that of the second SVEC.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_ge(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS bool AS 'MODULE_PATHNAME', 'svec_l2_ge' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

--! Returns a value indicating the relative values of the l2 norms of two SVECs.
--!
CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svec_l2_cmp(MADLIB_SCHEMA.svec,MADLIB_SCHEMA.svec) RETURNS integer AS 'MODULE_PATHNAME', 'svec_l2_cmp' LANGUAGE C IMMUTABLE
m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `NO SQL', `');

/*
DROP OPERATOR IF EXISTS < (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) CASCADE ;
DROP OPERATOR IF EXISTS <= (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) CASCADE ;
DROP OPERATOR IF EXISTS <> (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) ;
DROP OPERATOR IF EXISTS == (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) CASCADE ;
DROP OPERATOR IF EXISTS > (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) CASCADE ;
DROP OPERATOR IF EXISTS >= (MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec) CASCADE ;
DROP OPERATOR IF EXISTS *|| (int4, MADLIB_SCHEMA.svec) ;
*/

CREATE OPERATOR MADLIB_SCHEMA.< (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_lt,
	commutator = operator(MADLIB_SCHEMA.>) , negator = operator(MADLIB_SCHEMA.>=) ,
	restrict = scalarltsel, join = scalarltjoinsel
);
CREATE OPERATOR MADLIB_SCHEMA.<= (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_le,
	commutator = operator(MADLIB_SCHEMA.>=) , negator = operator(MADLIB_SCHEMA.>) ,
	restrict = scalarltsel, join = scalarltjoinsel
);
CREATE OPERATOR MADLIB_SCHEMA.<> (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_ne,
	commutator = operator(MADLIB_SCHEMA.<>) ,
	negator = operator(MADLIB_SCHEMA.==),
	restrict = eqsel, join = eqjoinsel
);
CREATE OPERATOR MADLIB_SCHEMA.== (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_eq,
	commutator = operator(MADLIB_SCHEMA.==) ,
	negator = operator(MADLIB_SCHEMA.<>) ,
	restrict = eqsel, join = eqjoinsel
);
CREATE OPERATOR MADLIB_SCHEMA.>= (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_ge,
	commutator = operator(MADLIB_SCHEMA.<=) , negator = operator(MADLIB_SCHEMA.<) ,
	restrict = scalargtsel, join = scalargtjoinsel
);
CREATE OPERATOR MADLIB_SCHEMA.> (
	leftarg = MADLIB_SCHEMA.svec, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_gt,
	commutator = operator(MADLIB_SCHEMA.<) , negator = operator(MADLIB_SCHEMA.<=) ,
	restrict = scalargtsel, join = scalargtjoinsel
);

CREATE OPERATOR MADLIB_SCHEMA.*|| (
	leftarg = int4, rightarg = MADLIB_SCHEMA.svec, procedure = MADLIB_SCHEMA.svec_concat_replicate
);

DROP OPERATOR CLASS IF EXISTS MADLIB_SCHEMA.svec_ops USING btree;

CREATE OPERATOR CLASS MADLIB_SCHEMA.svec_ops
DEFAULT FOR TYPE MADLIB_SCHEMA.svec USING btree AS
OPERATOR        1       MADLIB_SCHEMA.< ,
OPERATOR        2       MADLIB_SCHEMA.<= ,
OPERATOR        3       MADLIB_SCHEMA.== ,
OPERATOR        4       MADLIB_SCHEMA.>= ,
OPERATOR        5       MADLIB_SCHEMA.> ,
FUNCTION        1       MADLIB_SCHEMA.svec_cmp(MADLIB_SCHEMA.svec, MADLIB_SCHEMA.svec);