greeness

#The input data should be in CSV format.
# Each row is a campaign. The columns are (description and variable name):
# Campaign id (id)
# Conversion rate (cvr)
# Bid price (price)
# Max allocated campaign budget (max_allocation)

import csv
import sys

greeness

greeness

% Exploiting feature covariance in high-dimensional online learning
% http://videolectures.net/aistats2010_ma_efcih/
% http://sysnet.ucsd.edu/projects/url/

function [err,mu,sigma,mem] = cw(X,Y,params)
% X is k features by N instances
% Y is 1 label in {-1,1} by N instances
% params is struct containing options

% err: cumulative mistakes after each example

greeness

[Python Scientific Lecture Notes]: (http://scipy-lectures.github.io/index.html)

greeness

simple and clear algorithm description:
http://www.titouangalopin.com/blog/articles/2014/05/simhash-or-the-way-to-compare-quickly-two-datasets

a python impl
http://liangsun.org/posts/a-python-implementation-of-simhash-algorithm/

original paper of simhash
http://www.cs.princeton.edu/courses/archive/spr04/cos598B/bib/CharikarEstim.pdf

simhash impl doc

greeness

import json
from sys import stdout
from random import choice, seed, shuffle
from copy import deepcopy

basic_shapes = [
    [(0,0), (1,0), (2,0)], 
    [(0,0), (0,1), (0,2)],
    [(0,0), (-1,1), (-2,2)],
    [(0,0), (-1,0), (-2,0)],

greeness

http://users.ics.aalto.fi/ella/publications/randproj_kdd.pdf

greeness

import json
from datetime import datetime
from copy import deepcopy
from random import random

def append_features_from_user_block(user_block, user_row):
    user_row["num_turns"] = len(user_block)
    user_row["min_balance"] = min([x["balance"] for x in user_block])
    user_row["max_balance"] = max([x["balance"] for x in user_block])
    user_row["min_level"] = min([x["level"] for x in user_block])

greeness

from datetime import datetime
import numpy as np

age_ranges = ['13-24', '25-35', '36-52', '53-64']
device_options = ['Phone', 'Tablet']
platform_options = ['a', 'i']

def weighted_choice(weights):
    totals = np.cumsum(weights)
    norm = totals[-1]

greeness

{
 "metadata": {
  "name": "2d Convolution"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {