Skip to content

Instantly share code, notes, and snippets.

@olooney
olooney / xgboost_keras_mp.py
Created June 1, 2018 01:47
Compare XGBoost and Keras head-to-head with parallel processing
# MP setup *must* happen at the very start of the process!
# Otherwise you'll get errors or it will be ignored.
import multiprocessing
multiprocessing.set_start_method('forkserver')
import os
os.environ["OMP_NUM_THREADS"] = "10"
import numpy as np
import pandas as pd
import pickle
@olooney
olooney / pnorm.sql
Created June 13, 2018 23:28
PostgreSQL pnorm() function calculated the c.d.f. of the normal Gaussian distribution. This function match's R's build in pnorm() function to within +/- 2e-7 over the entire real line. However, it's a constant 1/0 above/below z=+7/-7.
CREATE OR REPLACE FUNCTION pnorm(z double precision) RETURNS double precision AS $$
SELECT CASE
WHEN $1 >= 0 THEN 1 - POWER(((((((0.000005383*$1+0.0000488906)*$1+0.0000380036)*$1+0.0032776263)*$1+0.0211410061)*$1+0.049867347)*$1+1),-16)/2
ELSE 1 - pnorm(-$1)
END;
$$ LANGUAGE SQL IMMUTABLE STRICT;
@olooney
olooney / sim.R
Created July 9, 2018 18:59 — forked from mikeguggis/sim.R
Stack Exchange
library(MASS)
library(ggplot2)
results <- NULL
# Generate simulated data
for ( seed in 1:30 ) {
set.seed(seed+42)
mu = c(0,-1+seed/10)
Sigma = matrix(c(1.5,-.5,-.5,.7),2,2)
@olooney
olooney / algorithm_u.py
Created July 11, 2018 23:44
Python 3 conversion of a Python 2 implementation of Knuth's "algorithm u" I found online
# Knuth's algorithm to partition ns into m sets.
# https://codereview.stackexchange.com/questions/1526/finding-all-k-subset-partitions
# http://cs.utsa.edu/~wagner/knuth/
# http://cs.utsa.edu/~wagner/knuth/fasc3b.pdf
# ns must be a list of integers
# m must be an integer less than len(ns)
# this is a generator expression
def algorithm_u(ns, m):
def visit(n, a):
@olooney
olooney / README.md
Last active January 26, 2019 14:44
SCRIPT-8
# https://www.youtube.com/watch?v=d3mHfqd0VZY
def harmonic(x, y, n=6, epsilon=1e-3):
r = sqrt(x/y)
for a in range(1, n):
for b in range(1, n):
if abs(r - a/b) < epsilon:
return True
return False
for x in np.arange(1.6, 6.4, 0.2):
X = np.hstack([np.ones(shape=(23, 1)), np.random.normal(size=(23, 2))])
Theta = np.array([0.5, +0.1, -0.2]).reshape( (3, 1) )
Y = X @ Theta + np.random.normal(0, 0.1, size=(23, 1))
Theta_hat = np.linalg.inv(X.T @ X) @ X.T @ Y
Theta_hat
from typing import NamedTuple, Any, Optional
class Node(NamedTuple):
"""A single Node in a binary tree."""
value: Any
left: Node
right: Node
def count(node: Optional[Node]) -> int:
"""Count the total number of nodes in a tree rooted at this node."""
@olooney
olooney / parallel_steps.py
Created October 25, 2019 00:20
Python parallel worker threads with back-pressure
import os
import sys
from time import sleep
from threading import Thread
from queue import Queue
import codecs
def hard_work(x):
for n in range(100001):
x = codecs.encode(x, 'rot_13')
from scipy.optimize import linear_sum_assignment
import numpy as np
def maximize_trace(X):
"""
Maximize the trace of a square matrix using only row and
column permutations. Furthermore, sort the trace
in desending order so that largest value ends
up the upper left and the smallest in the lower right.