Madlib PostgreSQL

apriori

CREATE TEMP TABLE test_data ON COMMIT DROP AS (SELECT
	warehouses.product_id::TEXT,
	user_id AS transaction_id
FROM
	"line_items"
INNER JOIN "users" ON "users"."id" = "line_items"."user_id"
INNER JOIN "warehouses" ON "warehouses"."id" = "line_items"."warehouse_id"
INNER JOIN "products" ON "products"."id" = "warehouses"."product_id"
AND (products.deleted_at IS NULL)
WHERE
	(
		line_items.deleted_at IS NULL
	)
AND (
	users.partner_type = 0
	AND products.is_quantity_supplier = 1
	OR products.storehouse_ids && ARRAY [ 1 ]
	AND line_items.pre_order_at IS NOT NULL
)
);

SELECT * FROM madlib.assoc_rules( 0.0003,
                                   0.001,
                                  'transaction_id',
                                  'product_id',
                                  'test_data',
                                  NULL,
                                  FALSE
                                );

SELECT * FROM assoc_rules
ORDER BY support DESC;

glm

-- http://doc.madlib.net/latest/group__grp__glm.html

DROP TABLE IF EXISTS glm_model, glm_model_summary, ads_regression, ads_regression_dummy;


CREATE TABLE ads_regression AS (
  SELECT id, price, price_m, repair_type::VARCHAR, wall_type::VARCHAR, ceiling_height::VARCHAR  FROM ads WHERE ad_type = 0  AND repair_type is not NULL and wall_type is not NULL limit 100
);

SELECT madlib.create_indicator_variables('ads_regression', 'ads_regression_dummy', 'wall_type,repair_type');

SELECT madlib.glm(  'ads_regression_dummy',
  'glm_model',
  'price_m',
  'ARRAY[1.0,"repair_type_3","wall_type_1"]',
  'family=poisson, link=identity'
);

SELECT * FROM glm_model;

SELECT
  ads_regression_dummy.id, 
  ads_regression_dummy.price_m,
  madlib.glm_predict_poisson(
    coef,
    ARRAY[1, "repair_type_3", "wall_type_1"]::float8[],
    'identity') AS poisson_count
FROM ads_regression_dummy, glm_model
ORDER BY ads_regression_dummy.id;

kmeans

CREATE TEMP TABLE ads_kmeans ON COMMIT DROP AS (
  SELECT id, ARRAY[price, repair_type] as points
  FROM ads WHERE ad_type = 0 AND repair_type is not NULL
);

SELECT
  ads_kmeans.*, 
  (madlib.closest_column(centroids, points)).column_id as cluster_id
FROM
  ads_kmeans,
  (
    SELECT *
    FROM madlib.kmeanspp(
      'ads_kmeans',
      'points',
      5,
      'madlib.squared_dist_norm2',
      'madlib.avg',
      20,
      0.001
    )
) as centroids

madlib_install.sh

#!/bin/bash

# https://github.com/madlib/madlib/wiki/Building-MADlib-from-Source
apt-get install -y cmake
apt-get install -y postgresql-server-dev-9.3
apt-get install -y postgresql-plpython-9.3
apt-get install -y libkrb5-dev

git clone https://github.com/madlib/madlib.git

cd madlib
./configure
cd build/
make
sudo make install

# gz - db name 
sudo -u postgres createlang plpythonu gz
/usr/local/madlib/bin/madpack -p postgres -c postgres@localhost/gz install

run command

wget --no-check-certificate https://gist.githubusercontent.com/ermolaev/6896e64d7fa121457d2c/raw/madlib_install.sh && sudo sh madlib_install.sh


https://github.com/pgRouting/pgrouting/issues/274

http://hssl.cs.jhu.edu/wiki/doku.php?id=cs423:schen:assignment3