mkoehrsen

select 'CREATE OR REPLACE VIEW dp_' || table_name ||
  E' AS\nSELECT ' || array_to_string(array_agg(format('%s',column_name)),E',\n       ') ||
  E'\nFROM ' || table_name || E';\n'
from information_schema.columns
where table_schema = 'public'
group by table_name;

mkoehrsen

with my_columns as 
  (select table_name,array_agg(json_build_object('column_name', column_name, 'is_nullable', is_nullable, 'data_type', data_type, 'ordinal_position', ordinal_position)) cols 
  from information_schema.columns 
  where table_schema='public' 
  group by table_name),
my_fks as 
  (select distinct r.constraint_name,
         k.table_name from_table,
         c.table_name to_table
  from information_schema.key_column_usage k

mkoehrsen

import numpy

def generate_max_zero_probabilities(maxZ):
    maxZ += 1
    row1 = numpy.array(list(map(lambda x: pow(2,-(x+1)),range(maxZ))))
    row1_cum = row1.cumsum() - row1 # non-inclusive cumsum
    yield (1,row1)

    prev_row = row1
    prev_row_cum = row1_cum

mkoehrsen

# Superpixel segmentation approach that seems to give pretty good contiguous segments.
# (SLIC and quickshift don't seem to guarantee contiguity). The approach is to get initial
# segments from SLIC, use the centroid of each as a marker for watershed, then clean up.

import os, argparse
from skimage import segmentation
from skimage.future import graph
import cv2, numpy
import tempfile
import random