lhk

import shelve

def save_scope(session_id='default'):
    filename = f'/tmp/{session_id}.out'
    shelf = shelve.open(filename, 'n')

    for name in dir():
        try:
            shelf[name] = globals()[name]
        except Exception as e:

lhk

  name: eval_env
channels:
  - defaults
dependencies:
  - _libgcc_mutex=0.1=main
  - asn1crypto=0.24.0=py36_0
  - attrs=19.1.0=py36_1
  - backcall=0.1.0=py36_0
  - blas=1.0=mkl
  - bleach=3.1.0=py36_0

lhk

-- a bit of subsampling, to make the size more manageable
-- I'm using the first N events, ordered by timestamp
-- this way, the learning traces should not be interrupted.
drop table if exists smaller;
create table smaller
as
select *
from log
order by timestamp
limit 20000000;

lhk

# coding: utf-8

import numpy as np
from scipy.stats import multivariate_normal

def metropolis_hastings(f, num_samples, burn_in, result_queue=None):

    # sample normal values as stepsize for the updates
    # important: g is symmetric, so we don't have to use it in the calculation of alpha below
    steps = np.random.normal(0, 1, (num_samples, 2))

lhk

import json
import requests


data = {"song_id": "sjrt56usng",
        "author_id": "spdfhu",
        "song_name": "best song ever",
        "artist_name": "onehitwonder",
        "voters": {

lhk

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",

lhk

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [

lhk

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [

lhk

lhk