Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

making Libra produce a JSON output #5

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

making Libra produce a JSON output #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
73 changes: 68 additions & 5 deletions src/libra/engine/backward.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
"""

import itertools
import json
import operator
import time
from copy import deepcopy
Expand Down Expand Up @@ -127,6 +128,8 @@ def __init__(self, cfg, manager, domain, semantics, specification, steps=None, m

self.cpu = cpu_count() if cpu is None else cpu

self.json = Manager().dict()

@property
def initial(self):
"""Initial analysis state
Expand Down Expand Up @@ -165,6 +168,13 @@ def feasibility(self, state, manager, disjuncts, key=None, chunk=None):
if len(outcomes) <= 1 and None not in outcomes:
classes = ', '.join(str(outcome) for outcome in outcomes)
print(Fore.GREEN + '✔︎ No Bias ({}) in {}'.format(classes, chunk), Style.RESET_ALL)
lock.acquire()
curr = self.json.get('fair', list())
curr.append({
'partition': self.chunk2json(chunk)
})
self.json['fair'] = curr
lock.release()
return True, list(), len(self.activations)
return feasible, patterns, _disjunctions

Expand Down Expand Up @@ -255,10 +265,10 @@ def worker1(self, id, color, queue1, manager):
queue1.put((None, None, None, None, None, None, None, None, None, None, None))
break
r_assumptions = '1-Hot: {}'.format(
', '.join('{}'.format('|'.join('{}'.format(var) for var in case)) for (case, _, _) in assumptions)
', '.join('{}'.format('∨'.join('{}'.format(var) for var in case)) for (case, _, _) in assumptions)
) if assumptions else ''
r_ranges = 'Ranges: {}'.format(
', '.join('{} ∈ [{}, {}]'.format(feature, lower, upper) for feature, (lower, upper) in ranges)
'; '.join('{} ∈ [{}, {}]'.format(feature, lower, upper) for feature, (lower, upper) in ranges)
)
r_partition = '{} | {}'.format(r_assumptions, r_ranges) if r_assumptions else '{}'.format(r_ranges)
print(color + r_partition, Style.RESET_ALL)
Expand Down Expand Up @@ -433,10 +443,40 @@ def worker1(self, id, color, queue1, manager):
if self.explored.value >= 100:
queue1.put((None, None, None, None, None, None, None, None, None, None, None))
found = '‼ Unchecked Bias in {}'.format(r_partition)
lock.acquire()
curr = self.json.get('unknown', list())
curr.append({
'partition': self.chunk2json(r_partition)
})
self.json['unknown'] = curr
lock.release()
print(Fore.RED + found, Style.RESET_ALL)
progress = 'Progress for #{}: {}% of {}% ({}% fair)'.format(id, self.feasible.value, self.explored.value, self.fair.value)
print(Fore.YELLOW + progress, Style.RESET_ALL)

def chunk2json(self, chunk):
chunks = chunk.split('|')
if len(chunks) > 1:
onehot = list(map(lambda x: x.split('∨'), chunks[0].strip().replace('1-Hot: ', '').split(', ')))
chunksranges = chunks[1].strip().replace('Ranges: ', '').split('; ')
else:
onehot = list()
chunksranges = chunks[0].strip().replace('Ranges: ', '').split('; ')
ranges = dict()
for chunkrange in chunksranges:
chunked = chunkrange.split(' ∈ ')
bounds = chunked[1].strip().replace('[', '').replace(']', '').split(', ')
ranges[chunked[0]] = {
'lower bound': float(bounds[0]),
'upper bound': float(bounds[1])
}
partition = {
'active 1-hot encoded inputs': onehot,
'ranges for continuous inputs': ranges,
'constraints': list()
}
return partition

def bias_check(self, chunk, result, ranges, percent):
"""Check for algorithmic bias

Expand Down Expand Up @@ -483,6 +523,22 @@ def bias_check(self, chunk, result, ranges, percent):
biases.add(representation)
pair = '{}->{} vs {}->{}'.format(sensitive1, outcome1, sensitive2, outcome2)
found = '✘ Bias Found ({})! in {}:\n{}'.format(pair, chunk, representation)
counterexample = {
'sensitive1': str(sensitive1),
'outcome1': str(outcome1),
'sensitive2': str(sensitive2),
'outcome2': str(outcome2)
}
partition = self.chunk2json(chunk)
partition['constraints'] = representation.split(' ∧ ')
lock.acquire()
curr = self.json.get('biased', list())
curr.append({
'counterexample': counterexample,
'partition': partition,
})
self.json['biased'] = curr
lock.release()
print(Fore.RED + found, Style.RESET_ALL)
if nobias:
outcomes = set()
Expand All @@ -502,6 +558,13 @@ def bias_check(self, chunk, result, ranges, percent):
outcomes.add(outcome)
classes = ', '.join(str(outcome) for outcome in outcomes)
print(Fore.GREEN + '✔︎ No Bias ({}) in {}'.format(classes, chunk), Style.RESET_ALL)
lock.acquire()
curr = self.json.get('fair', list())
curr.append({
'partition': self.chunk2json(chunk)
})
self.json['fair'] = curr
lock.release()
else:
total_size = 1
for _, (lower, upper) in ranges:
Expand Down Expand Up @@ -613,10 +676,10 @@ def worker2(self, id, color, queue2, manager, total):
# check for bias
for assumptions, unpacked, ranges, percent in pack:
r_assumptions = '1-Hot: {}'.format(
', '.join('{}'.format('|'.join('{}'.format(var) for var in case)) for (case, _, _) in assumptions)
', '.join('{}'.format('∨'.join('{}'.format(var) for var in case)) for (case, _, _) in assumptions)
) if assumptions else ''
r_ranges = 'Ranges: {}'.format(
', '.join('{} ∈ [{}, {}]'.format(feature, lower, upper) for feature, (lower, upper) in ranges)
'; '.join('{} ∈ [{}, {}]'.format(feature, lower, upper) for feature, (lower, upper) in ranges)
)
r_partition = '{} | {}'.format(r_assumptions, r_ranges) if r_assumptions else '{}'.format(r_ranges)
if unpacked:
Expand Down Expand Up @@ -868,7 +931,7 @@ def analyze(self, initial, inputs=None, outputs=None, activations=None, analysis
else:
log = '{} ({}% certified in the pre-analysis) {}s'.format(self.feasible.value, self.fair.value, end1 - start1)
print('\nDone!')
return log
return self.json.copy()


class BiasBackwardSemantics(DefaultBackwardSemantics):
Expand Down
9 changes: 7 additions & 2 deletions src/libra/engine/bias_analysis.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,8 @@
"""
import ast
import ctypes
import json
import os
import time
from enum import Enum
from queue import Queue
Expand Down Expand Up @@ -203,11 +205,14 @@ def main(self, path):
r_vars.append(PyVar(variable.name))
environment = PyEnvironment([], r_vars)
self.activations, self.splits, self.relus = self.lyra2apron(environment)
self.run()
result = self.run()
with open(os.path.basename(self.path).replace('.py','.json'), 'w', encoding='utf8') as outfile:
json.dump(result, outfile, indent=4, separators=(',', ':'), ensure_ascii=False)

def run(self):
start = time.time()
log = self.interpreter().analyze(self.state(), inputs=self.inputs, outputs=self.outputs, activations=self.activations, analysis=self.analysis)
result = self.interpreter().analyze(self.state(), inputs=self.inputs, outputs=self.outputs, activations=self.activations, analysis=self.analysis)
end = time.time()
print('Total Time: {}s'.format(end - start), Style.RESET_ALL)
# print('{} {}s'.format(log, end - start))
return result