def get_jaccard_distribution(pf, args, start_t):
"""Samples random pairs and saves their Jaccard similarity to CSV.
The result is used by jaccard_distribution.R to fit an
distribution to the data.
"""
csv = CsvWriter(args.results, append=True)
csv.write_header(['u1', 'u2', 'jac_sim', 'sig_sim'])
for i, (u1, u2) in enumerate(
zip(np.random.permutation(pf.n_docs),
np.random.permutation(pf.n_docs))):
if u1 == u2: next
print("Wrote {} similarities".format(i), end='\r')
csv.write([u1, u2, pf.jaccard_similarity(u1, u2), pf.sig_sim(u1, u2)])
return True
def get_candidate_distribution(pf, args, start_t):
"""Samples pairs from candidates, stratified by signature similarity.
The result is stored in a CSV, including the signature similarity,
Jaccard similarity, and weight (> 1 if there were more samples in
the stratum than max_per_step, to get approximately accurate
frequencies).
"""
candidates = list(pf.candidates())
csv_file = args.results
csv = CsvWriter(csv_file, append=True)
csv.write_header([
'run_id', 'u1', 'u2', 'sig_sim', 'jac_sim', 'sig_len', 'bands',
'max_buckets', 'used_buckets', 'weight'
])
run_id = datetime.now().isoformat()
lim = 0
step = 0.05
max_per_step = 100
while lim < 1:
cand = [(c, sim) for c, sim in candidates
if sim >= lim and sim < lim + step]
n = len(cand)
print("{} candidates between {} and {}".format(n, lim, lim + step))
weight = max(n, max_per_step) / max_per_step
for i in np.random.permutation(n)[:max_per_step]:
(c1, c2), sim = cand[i]
csv.write([
run_id, c1, c2, sim,
pf.jaccard_similarity(c1, c2), pf.sig_len, pf.n_bands,
pf.max_buckets,
len(pf.buckets), weight
])
lim += step
return True
import time
import subprocess
import os
import gc
import data
from csv_writer import CsvWriter
from util import ensure_directory
"""
This file simulates the evaluation environment and stores results
as a CSV.
"""
csv = CsvWriter("diagnostics/out/evaluation.csv", append=True)
csv.write_header([
'note', 'batch', 'run', 'found', 'incorrect', 'time', 'ppm', 'terminated'
])
def jaccard_sim(data, u1, u2):
m1 = data.movie[data.user == u1]
m2 = data.movie[data.user == u2]
return len(np.intersect1d(m1, m2)) / len(np.union1d(m1, m2))
def run_evaluation(note, batch=0, runs=5):
for run in range(runs):
cmd = [
"python3",
"main.py",
"--rows",
import numpy as np
import multiprocessing as mp
from sklearn.model_selection import ParameterGrid
from datetime import datetime
import time
import subprocess
import os
from csv_writer import CsvWriter
from util import ensure_directory
csv = CsvWriter("diagnostics/out/experiments.csv", append=True)
csv.write_header(
['batch_id', 'run_id', 'bands', 'rows', 'max_buckets', 'time', 'count'])
batch_id = datetime.now().isoformat()
def count_lines_in_file(filename):
if not os.path.exists(filename):
return 0
count = 0
with open(filename) as f:
for l in f:
if l.strip() != '':
count += 1
return count
def perform(params):