Skip to content

Commit

Permalink
Merge pull request #90 from fplll/improve_pruner
Browse files Browse the repository at this point in the history 
Improve pruner
  • Loading branch information
@malb
malb committed Jul 29, 2017
2 parents 5f6c4c3 + da35697 commit 7b061fb
Show file tree
Hide file tree
Showing 11 changed files with 379 additions and 512 deletions.
2 changes: 1 addition & 1 deletion src/fpylll/__init__.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
from .fplll.bkz_param import load_strategies_json
from .fplll.svpcvp import SVP
from .fplll.svpcvp import CVP
from .fplll.pruner import prune, Pruner
from .fplll.pruner import Pruning
from .fplll.sieve_gauss import GaussSieve
from .util import ReductionError
from .util import set_random_seed, set_precision, get_precision
Expand Down
12 changes: 1 addition & 11 deletions src/fpylll/fplll/bkz_param.pxd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,20 +4,10 @@ from libcpp.vector cimport vector
from decl cimport bkz_auto_abort_core_t, fplll_gso_type_t
from gso cimport MatGSO
from fplll cimport BKZParam as BKZParam_c
from fplll cimport Pruning as Pruning_c
from fplll cimport PruningParams as PruningParams_c
from fplll cimport Strategy as Strategy_c
from fplll cimport PrunerMetric

cdef class Pruning:
cdef Pruning_c _core

@staticmethod
cdef Pruning from_cxx(Pruning_c & p)

@staticmethod
cdef to_cxx(Pruning_c& self, Pruning p)


cdef class Strategy:
cdef Strategy_c _core

Expand Down
208 changes: 24 additions & 184 deletions src/fpylll/fplll/bkz_param.pyx
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,184 +18,19 @@ from fplll cimport BKZ_DEF_RERANDOMIZATION_DENSITY
from fplll cimport PRUNER_METRIC_PROBABILITY_OF_SHORTEST
from fplll cimport PRUNER_METRIC_EXPECTED_SOLUTIONS
from fplll cimport LLL_DEF_DELTA
from fplll cimport Pruning as Pruning_c
from fplll cimport PruningParams as PruningParams_c
from fplll cimport Strategy as Strategy_c
from fplll cimport load_strategies_json as load_strategies_json_c
from fplll cimport strategy_full_path

from fpylll.util cimport check_delta, check_pruner_metric
from cython.operator cimport dereference as deref, preincrement as inc

from pruner cimport PruningParams

from collections import OrderedDict
import json

cdef class Pruning:
"""
Pruning parameters.
"""
def __init__(self, radius_factor, coefficients, expectation=1.0,
metric="probability", detailed_cost=tuple()):
"""Create new pruning parameters object.
:param radius_factor: ratio of radius to Gaussian heuristic
:param coefficients: a list of pruning coefficients
:param expectation: success probability or number of solutions
:param metric: either "probability" or "solutions"
"""
if radius_factor <= 0:
raise ValueError("Radius factor must be > 0")

cdef PrunerMetric met = <PrunerMetric>check_pruner_metric(metric)

if met == PRUNER_METRIC_PROBABILITY_OF_SHORTEST:
if expectation <= 0 or expectation > 1:
raise ValueError("Probability must be between 0 and 1")

self._core.radius_factor = radius_factor
self._core.expectation = expectation
self._core.metric = met

for c in coefficients:
self._core.coefficients.push_back(c)

for c in detailed_cost:
self._core.detailed_cost.push_back(c)

@staticmethod
cdef Pruning from_cxx(Pruning_c& p):
"""
Load Pruning object from C++ Pruning object.
.. note::
All data is copied, i.e. `p` can be safely deleted after this function returned.
"""

cdef Pruning self = Pruning(1.0, ())
self._core = p
return self

@staticmethod
cdef to_cxx(Pruning_c& self, Pruning p):
"""
Store pruning object in C++ pruning object.
.. note::
All data is copied, i.e. `p` can be safely deleted after this function returned.
"""
self.radius_factor = p._core.radius_factor
self.expectation = p._core.expectation
self.metric = p._core.metric
for c in p._core.coefficients:
self.coefficients.push_back(c)
for c in p._core.detailed_cost:
self.detailed_cost.push_back(c)

@staticmethod
def LinearPruning(block_size, level):
"""
Set all pruning coefficients to 1, except the last <level>
coefficients, these will be linearly with slope `-1 / block_size`.
:param block_size: block size
:param level: level
"""
sig_on()
cdef Pruning_c p = Pruning_c.LinearPruning(block_size, level)
sig_off()
return Pruning.from_cxx(p)

def __reduce__(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> import pickle
>>> print(pickle.loads(pickle.dumps(Pruning(1.0, [1.0, 0.6, 0.3], 1.0))))
Pruning<1.000000, (1.00,...,0.30), 1.0000>
"""
return Pruning, (self.radius_factor, self.coefficients, self.expectation, self.metric, self.detailed_cost)

def __str__(self):
return "Pruning<%f, (%.2f,...,%.2f), %.4f>"%(self.radius_factor, self.coefficients[0], self.coefficients[-1], self.expectation)

@property
def radius_factor(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> pr = Pruning(1.0, [1.0, 0.6, 0.3], 0.9)
>>> pr.radius_factor
1.0
"""
return self._core.radius_factor

@property
def expectation(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> pr = Pruning(1.0, [1.0, 0.6, 0.3], 0.9)
>>> pr.expectation
0.9
"""
return self._core.expectation

@property
def metric(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> pr = Pruning(1.0, [1.0, 0.6, 0.3], 0.9)
>>> pr.metric
'probability'
"""
if self._core.metric == PRUNER_METRIC_PROBABILITY_OF_SHORTEST:
return "probability"
elif self._core.metric == PRUNER_METRIC_EXPECTED_SOLUTIONS:
return "solutions"
else:
raise NotImplementedError("Metric %d not understood"%self._core.metric)

@property
def coefficients(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> pr = Pruning(1.0, [1.0, 0.6, 0.3], 0.9)
>>> pr.coefficients
(1.0, 0.6, 0.3)
"""
cdef list coefficients = []
cdef vector[double].iterator it = self._core.coefficients.begin()
while it != self._core.coefficients.end():
coefficients.append(deref(it))
inc(it)
return tuple(coefficients)

@property
def detailed_cost(self):
"""
>>> from fpylll.fplll.bkz_param import Pruning
>>> pr = Pruning(1.0, [1.0, 0.6, 0.3], 0.9)
>>> pr.detailed_cost
()
"""
cdef list detailed_cost = []
cdef vector[double].iterator it = self._core.detailed_cost.begin()
while it != self._core.detailed_cost.end():
detailed_cost.append(deref(it))
inc(it)
return tuple(detailed_cost)


cdef class Strategy:
"""
A strategy is a collection of pruning coefficients for a variety
Expand All @@ -215,13 +50,13 @@ cdef class Strategy:
self._core.block_size = block_size

for p in pruning_parameters:
if not isinstance(p, Pruning):
p = Pruning(*p)
self._core.pruning_parameters.push_back((<Pruning>p)._core)
if not isinstance(p, PruningParams):
p = PruningParams(*p)
self._core.pruning_parameters.push_back((<PruningParams>p)._core)

if len(pruning_parameters) == 0:
p = Pruning(1.0, [1.0 for _ in range(self.block_size)], 1.0)
self._core.pruning_parameters.push_back((<Pruning>p)._core)
p = PruningParams(1.0, [1.0 for _ in range(self.block_size)], 1.0)
self._core.pruning_parameters.push_back((<PruningParams>p)._core)

for p in preprocessing_block_sizes:
if p<=2:
Expand All @@ -241,9 +76,9 @@ cdef class Strategy:
closest_dist = 2**80
best = None
for pruning in self.pruning_parameters:
if abs(pruning.radius_factor - gh_factor) < closest_dist:
if abs(pruning.gh_factor - gh_factor) < closest_dist:
best = pruning
closest_dist = abs(pruning.radius_factor - gh_factor)
closest_dist = abs(pruning.gh_factor - gh_factor)
assert(best is not None)
return best

Expand All @@ -260,7 +95,7 @@ cdef class Strategy:
d = OrderedDict()
d["block_size"] = self.block_size
d["preprocessing_block_sizes"] = self.preprocessing_block_sizes
d["pruning_parameters"] = tuple([(p.radius_factor, p.coefficients, p.expectation, p.metric, p.detailed_cost)
d["pruning_parameters"] = tuple([(p.gh_factor, p.coefficients, p.expectation, p.metric, p.detailed_cost)
for p in self.pruning_parameters])
return d

Expand All @@ -275,9 +110,11 @@ cdef class Strategy:

def __reduce__(self):
"""
>>> from fpylll import Pruning
>>> from fpylll.fplll.bkz_param import Strategy
>>> import pickle
>>> print(pickle.loads(pickle.dumps(Strategy(20, [10], [Pruning(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)]))))
>>> p = Pruning.PruningParams(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)
>>> print(pickle.loads(pickle.dumps(Strategy(20, [10], [p]))))
Strategy< 20, (10), 0.50-0.50>
"""
Expand All @@ -293,7 +130,7 @@ cdef class Strategy:
cdef to_cxx(Strategy_c& self, Strategy s):

for p in s.pruning_parameters:
self.pruning_parameters.push_back((<Pruning>p)._core)
self.pruning_parameters.push_back((<PruningParams>p)._core)

for p in s.preprocessing_block_sizes:
self.preprocessing_block_sizes.push_back(p)
Expand All @@ -303,8 +140,9 @@ cdef class Strategy:
@property
def block_size(self):
"""
>>> from fpylll import Pruning
>>> from fpylll.fplll.bkz_param import Strategy
>>> s = Strategy(20, [10], [Pruning(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> s = Strategy(20, [10], [Pruning.PruningParams(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> s.block_size
20
Expand All @@ -314,8 +152,9 @@ cdef class Strategy:
@property
def preprocessing_block_sizes(self):
"""
>>> from fpylll import Pruning
>>> from fpylll.fplll.bkz_param import Strategy
>>> s = Strategy(20, [10], [Pruning(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> s = Strategy(20, [10], [Pruning.PruningParams(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> s.preprocessing_block_sizes
(10,)
Expand All @@ -330,16 +169,17 @@ cdef class Strategy:
@property
def pruning_parameters(self):
"""
>>> from fpylll import Pruning
>>> from fpylll.fplll.bkz_param import Strategy
>>> s = Strategy(20, [10], [Pruning(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> s = Strategy(20, [10], [Pruning.PruningParams(1.0, [1.0, 0.75, 0.5, 0.25], 0.5)])
>>> print(s.pruning_parameters[0])
Pruning<1.000000, (1.00,...,0.25), 0.5000>
PruningParams<1.000000, (1.00,...,0.25), 0.5000>
"""
cdef list pruning_parameters = []
cdef vector[Pruning_c].iterator it = self._core.pruning_parameters.begin()
cdef vector[PruningParams_c].iterator it = self._core.pruning_parameters.begin()
while it != self._core.pruning_parameters.end():
pruning_parameters.append(Pruning.from_cxx(deref(it)))
pruning_parameters.append(PruningParams.from_cxx(deref(it)))
inc(it)
return tuple(pruning_parameters)

Expand Down

0 comments on commit 7b061fb

Please sign in to comment.