arogozhnikov

Notes on edgedb

I’ve made a test evaluation of edgedb for the purposes of a small company.

My order of priorities (starting from most important):

1. correctness of results (data and code)
2. developer convenience
3. minimal maintenance
4. efficiency/scalability

arogozhnikov

arogozhnikov

# Run command with the same user ID as current user
# -v $(pwd):/tmp/mount - mount current directory to /tmp/mount/
# --env HOME="/tmp/" - some commands may need to be able to write to your home, se it to temporary folder
docker run -ti --rm -v $(pwd):/tmp/mount —user=$(id -u) --env HOME="/tmp/" debian:jessie

# Mount current users and group and be able to use them
# mount /etc/group and /etc/passwd read only
# set user from $USER
docker run -ti --rm -v $(pwd):/tmp/mount -w /tmp/hx -v /etc/group:/etc/group:ro -v /etc/passwd:/etc/passwd:ro —user=$USER debian:jessie

arogozhnikov

arogozhnikov

----> call function __getitem__
------> call function __contains__
<------ exit function __contains__
------> call function _getitem_column
--------> call function _get_item_cache
----------> call function get
------------> call function _get_items
<------------ exit function _get_items
------------> call function isnull
--------------> call function _isnull_new

arogozhnikov

# to be executed from different notebook!
def check_notebook_reproducibility(filename):
    md5s = []
    for attempt in [1, 2]:
        !jupyter nbconvert --to notebook --inplace --execute $filename --ExecutePreprocessor.timeout=86400
        result = !md5sum $filename
        md5s.append(result)
        print result
    a, b = md5s
    assert a == b, 'hashes are different'

arogozhnikov

def compile_fortran(source, modulename, extra_args=''):
    import tempfile
    import sys
    import numpy.f2py
    from numpy.distutils.exec_command import exec_command
    
    try:
        f = tempfile.NamedTemporaryFile(suffix='.f90')
        f.write(source)
        f.flush()

arogozhnikov

def isotonic_transform(x, y, alpha=0.001):
    target = y + 0.
    output = numpy.zeros(len(x))
    steps = 2 ** numpy.arange(int(numpy.log2(len(x)) - 1))[::-1]
    print steps
    for i in range(100):
        for step in steps:
            penalty = 2 * (target - output)
            diffs = - (output[step:] - output[:-step])
            diffs = diffs.clip(0) 

arogozhnikov

import numpy
from scipy.special import expit

class GRBM(object):
    def __init__(self, n_vis, n_hid, std=0.1):
        self.std = std
        self.W = numpy.random.normal(loc=-1./n_hid, scale=0.1, size=[n_vis + 1, n_hid + 1])
    
    def hidden_p(self, data):
        result = expit(data.dot(self.W))

arogozhnikov