def answers_generator(filename_gen, truth, epoch_from_filename=False,
use_shortname=True):
for fn in filename_gen:
answers = read_answers_file(fn)
if use_shortname:
fn = get_shortname(fn, epoch_from_filename)
yield (fn, answers)
results = defaultdict(list)
def answers_generator(filename_gen,
truth,
epoch_from_filename=False,
use_shortname=True):
for fn in filename_gen:
answers = read_answers_file(fn)
if use_shortname:
fn = get_shortname(fn, epoch_from_filename)
yield (fn, answers)
results = defaultdict(list)
def generate_ensembles(filename_gen,
ensemble_size,
truth,
sample_size=14,
replace=False,
randomise=False,
epoch_from_filename=False,
cat1_centre=None,
cat1_radius=0,
include=always,
exclude=never,
iterations=1):
answers_gen = answers_generator(filename_gen, truth, epoch_from_filename,
False)
scores = get_sorted_scores(answers_gen,
truth,
cat1_centre,
cat1_radius,
exclude,
epoch_from_filename=epoch_from_filename)
results = []
for i in range(iterations):
if randomise:
random.shuffle(scores)
cutoff = sample_size
singles = {}
essentials = set()
single_lines = []
if include is not always:
for c, fn, shortname in scores:
if include(shortname):
singles[shortname] = read_answers_file(fn)
cutoff -= 1
single_lines.append((c[0], shortname))
essentials.add(shortname)
if cutoff < 0:
cutoff = 0
for c, fn, shortname in scores[:cutoff]:
singles[shortname] = read_answers_file(fn)
single_lines.append((c[0], shortname))
ensembles = []
if replace:
combos = itertools.combinations_with_replacement
else:
combos = itertools.combinations
for names in combos(singles.keys(), ensemble_size):
ensemble = {}
if essentials and not essentials.intersection(names):
continue
for n in names:
answers = singles.get(n)
for k, v in answers.items():
score = ensemble.get(k, 0.0)
ensemble[k] = score + v
for k, v in ensemble.items():
ensemble[k] = v / ensemble_size
if cat1_centre is None:
centre = search_for_centre(ensemble, truth)
else:
centre = evaluate_fixed_cat1(ensemble, truth, cat1_centre,
cat1_radius)
ensembles.append((centre[0], names))
ensembles.sort()
results.append((single_lines, singles, ensembles))
return results
def generate_ensembles(filename_gen, ensemble_size, truth,
sample_size=14, replace=False, randomise=False,
epoch_from_filename=False,
cat1_centre=None, cat1_radius=0,
include=always, exclude=never, iterations=1):
answers_gen = answers_generator(filename_gen, truth,
epoch_from_filename, False)
scores = get_sorted_scores(answers_gen, truth, cat1_centre, cat1_radius,
exclude, epoch_from_filename=epoch_from_filename)
results = []
for i in range(iterations):
if randomise:
random.shuffle(scores)
cutoff = sample_size
singles = {}
essentials = set()
single_lines = []
if include is not always:
for c, fn, shortname in scores:
if include(shortname):
singles[shortname] = read_answers_file(fn)
cutoff -= 1
single_lines.append((c[0], shortname))
essentials.add(shortname)
if cutoff < 0:
cutoff = 0
for c, fn, shortname in scores[:cutoff]:
singles[shortname] = read_answers_file(fn)
single_lines.append((c[0], shortname))
ensembles = []
if replace:
combos = itertools.combinations_with_replacement
else:
combos = itertools.combinations
for names in combos(singles.keys(), ensemble_size):
ensemble = {}
if essentials and not essentials.intersection(names):
continue
for n in names:
answers = singles.get(n)
for k, v in answers.items():
score = ensemble.get(k, 0.0)
ensemble[k] = score + v
for k, v in ensemble.items():
ensemble[k] = v / ensemble_size
if cat1_centre is None:
centre = search_for_centre(ensemble, truth)
else:
centre = evaluate_fixed_cat1(ensemble, truth, cat1_centre,
cat1_radius)
ensembles.append((centre[0], names))
ensembles.sort()
results.append((single_lines, singles, ensembles))
return results
