def print_debug_data(self):
"""
Show weights statistics
"""
if not self.logger.isEnabledFor(logging.DEBUG):
return
self.weights.map_read()
self.bias.map_read()
self.gradient_bias.map_read()
self.gradient_weights.map_read()
weights = self.weights.mem
bias = self.bias.mem
grad_weights = self.gradient_weights.mem
grad_bias = self.gradient_bias.mem
weight_table = PrettyTable("TYPE", "Mean", "StdDev", "Min", "Max")
weight_table.float_format = ".10"
for (w_name, w_array) in [
("Weight", weights),
("Bias", bias),
("Grad Weight", grad_weights),
("Grad Bias", grad_bias),
]:
w_mean = w_stddev = w_min = w_max = None
if w_array is not None and w_array.size > 0:
w_mean = numpy.mean(w_array)
w_stddev = numpy.std(w_array)
w_min = numpy.min(w_array)
w_max = numpy.max(w_array)
weight_table.add_row(w_name, w_mean, w_stddev, w_min, w_max)
self.debug("\n" + weight_table.get_string())
def print_table(diffs):
table = PrettyTable("Unit", "Attribute", "Average Relative Diff", "Average Diff", "Max Diff")
for fn in table.field_names[:2]:
table.align[fn] = "l"
for fn in table.field_names[2:]:
table.align[fn] = "c"
for diff in diffs:
table.add_row(*diff[1:])
print(table)
def run(self):
self.input.map_read()
mem = self.input.mem[0]
labels = [(v, i) for i, v in enumerate(mem)]
labels.sort(reverse=True)
table = PrettyTable("label", "value")
table.float_format = ".5"
for row in labels[:self.top_number]:
table.add_row(*reversed(row))
self.info("Results:\n%s", table)
self.info("Max to mean ratio: %.1f", numpy.max(mem) / numpy.mean(mem))
def finished(response):
self.debug("Received %s", response)
try:
response = json.loads(response.decode('UTF-8'))
except ValueError as e:
self.exception("Failed to parse the response from server: %s",
response)
self.stop(Failure(e), False)
return
table = PrettyTable("Name", "Description", "Author", "Version",
"Date")
table.align["Name"] = table.align["Description"] = 'l'
for item in response:
table.add_row(*item)
print(table)
sys.stdout.flush()
self.stop()
def _print_label_stats(self, stats, set_name):
values = list(stats.values())
if sum(values) == 0:
self.info("No %s labels specified", set_name)
return
mean = int(numpy.mean(values))
stddev = int(numpy.std(values))
lmin = numpy.min(values)
amin = list(stats.keys())[numpy.argmin(values)]
lmax = numpy.max(values)
amax = list(stats.keys())[numpy.argmax(values)]
if has_colors() and stddev > mean / 10:
endstr = "\033[0m"
if stddev > mean / 2:
openstr = "\033[1;31m" # red
else:
openstr = "\033[1;33m" # yellow
else:
openstr = endstr = ""
self.info(
u"%s label cardinalities: min: %d (\"%s\"), max: %d (\"%s\"), avg:"
u" %d, %sσ: %d (%d%%)%s", set_name, lmin, amin, lmax, amax, mean,
openstr, stddev, stddev * 100 // mean, endstr)
if not self.logger.isEnabledFor(logging.DEBUG):
return
total = sum(values)
table = PrettyTable("Label", "Cardinality", "%", "Histogram")
table.align["Cardinality"] = "r"
table.align["%"] = "r"
table.align["Histogram"] = "l"
for k, v in stats.items():
table.add_row(k, v, "%.1f" % (v * 100 / total),
"*" * (v * 25 // lmax))
self.debug("Detailed %s label stats:\n%s", set_name, table)
def _print_label_stats(self, stats, set_name):
values = list(stats.values())
if sum(values) == 0:
self.info("No %s labels specified", set_name)
return
mean = int(numpy.mean(values))
stddev = int(numpy.std(values))
lmin = numpy.min(values)
amin = list(stats.keys())[numpy.argmin(values)]
lmax = numpy.max(values)
amax = list(stats.keys())[numpy.argmax(values)]
if has_colors() and stddev > mean / 10:
endstr = "\033[0m"
if stddev > mean / 2:
openstr = "\033[1;31m" # red
else:
openstr = "\033[1;33m" # yellow
else:
openstr = endstr = ""
self.info(
u"%s label cardinalities: min: %d (\"%s\"), max: %d (\"%s\"), avg:"
u" %d, %sσ: %d (%d%%)%s", set_name, lmin, amin, lmax, amax, mean,
openstr, stddev, stddev * 100 // mean, endstr)
if not self.logger.isEnabledFor(logging.DEBUG):
return
total = sum(values)
table = PrettyTable("Label", "Cardinality", "%", "Histogram")
table.align["Cardinality"] = "r"
table.align["%"] = "r"
table.align["Histogram"] = "l"
for k, v in stats.items():
table.add_row(k, v, "%.1f" % (v * 100 / total),
"*" * (v * 25 // lmax))
self.debug("Detailed %s label stats:\n%s", set_name, table)
def print_debug_data(self):
"""
Show weights statistics
"""
if not self.logger.isEnabledFor(logging.DEBUG):
return
self.weights.map_read()
self.bias.map_read()
self.gradient_bias.map_read()
self.gradient_weights.map_read()
weights = self.weights.mem
bias = self.bias.mem
grad_weights = self.gradient_weights.mem
grad_bias = self.gradient_bias.mem
weight_table = PrettyTable("TYPE", "Mean", "StdDev", "Min", "Max")
weight_table.float_format = ".10"
for (w_name, w_array) in [("Weight", weights), ("Bias", bias),
("Grad Weight", grad_weights),
("Grad Bias", grad_bias)]:
w_mean = w_stddev = w_min = w_max = None
if w_array is not None and w_array.size > 0:
w_mean = numpy.mean(w_array)
w_stddev = numpy.std(w_array)
w_min = numpy.min(w_array)
w_max = numpy.max(w_array)
weight_table.add_row(w_name, w_mean, w_stddev, w_min, w_max)
self.debug("\n" + weight_table.get_string())
def check_snapshot_size(self, size):
if size > self.SIZE_WARNING_THRESHOLD and self._warn_about_size:
self._warn_about_size = False
psizes = []
try:
for unit in self.workflow:
unit.stripped_pickle = True
psize = len(pickle.dumps(unit, protocol=4))
psizes.append((psize, unit))
unit.stripped_pickle = False
except:
self.warning("The snapshot size looks too big: %d bytes", size)
return
import gc
gc.collect()
psizes.sort(reverse=True)
pstable = PrettyTable("Unit", "Size")
pstable.align["Unit"] = "l"
for size, unit in psizes[:5]:
pstable.add_row(str(unit), size)
self.warning(
"The snapshot size looks too big: %d bytes. Here are top 5 "
"big units:\n%s", size, pstable)
def check_snapshot_size(self, size):
if size > self.SIZE_WARNING_THRESHOLD and self._warn_about_size:
self._warn_about_size = False
psizes = []
try:
for unit in self.workflow:
unit.stripped_pickle = True
psize = len(pickle.dumps(unit, protocol=4))
psizes.append((psize, unit))
unit.stripped_pickle = False
except:
self.warning("The snapshot size looks too big: %d bytes", size)
return
import gc
gc.collect()
psizes.sort(reverse=True)
pstable = PrettyTable("Unit", "Size")
pstable.align["Unit"] = "l"
for size, unit in psizes[:5]:
pstable.add_row(str(unit), size)
self.warning(
"The snapshot size looks too big: %d bytes. Here are top 5 "
"big units:\n%s", size, pstable)
def print_table(diffs):
table = PrettyTable("Unit", "Attribute", "Average Relative Diff",
"Average Diff", "Max Diff")
for fn in table.field_names[:2]:
table.align[fn] = "l"
for fn in table.field_names[2:]:
table.align[fn] = "c"
for diff in diffs:
table.add_row(*diff[1:])
print(table)
def print_stats(self):
if self.zmq_connection is None:
return
table = PrettyTable("", "receive", "send")
table.align[""] = "r"
table.add_row(
"all",
datetime.timedelta(seconds=self.zmq_connection.total_receive_time),
datetime.timedelta(seconds=self.zmq_connection.total_request_time))
try:
table.add_row(
"avg",
datetime.timedelta(seconds=self.zmq_connection.receive_timing),
datetime.timedelta(
seconds=self.zmq_connection.request_timings["update"]))
except KeyError:
pass
self.info("Timings:\n%s", table)
def print_stats(self, by_name=False, top_number=5):
"""Outputs various time statistics gathered with run_timed and
method_timed.
"""
stats = self.get_unit_run_time_stats(by_name)
time_all = sum(s[1] for s in stats)
if time_all > 0:
table = PrettyTable("#", "%", "time", "unit")
table.align["unit"] = "l"
top_time = 0
for i in range(1, min(top_number, len(stats)) + 1):
top_time += stats[i - 1][1]
table.add_row(i, int(stats[i - 1][1] * 100 / time_all),
datetime.timedelta(seconds=stats[i - 1][1]),
stats[i - 1][0])
table.add_row(u"Σ", int(top_time * 100 / time_all),
datetime.timedelta(seconds=top_time), "Top 5")
self.info(u"Unit run time statistics top:\n%s", table)
table = PrettyTable("units", "real", u"η,%")
table.add_row(datetime.timedelta(seconds=time_all),
datetime.timedelta(seconds=self._run_time_),
int(time_all * 100 / (self._run_time_ or 1)))
self.info(u"Total run time:\n%s", table)
table = PrettyTable("method", "%", "time")
table.align["method"] = "l"
time_all = 0
for k, v in sorted(self._method_time_.items()):
if k == "run":
continue
time_all += v
if self._run_time_ > 0:
table.add_row(k, int(v * 100 / self._run_time_),
datetime.timedelta(seconds=v))
if self.is_slave:
table.add_row(u"Σ", int(time_all * 100 / self._run_time_),
datetime.timedelta(seconds=time_all))
if time_all > 0:
self.info(u"Workflow methods run time:\n%s", table)
def print_stats(self, by_name=False, top_number=5):
"""Outputs various time statistics gathered with run_timed and
method_timed.
"""
stats = self.get_unit_run_time_stats(by_name)
time_all = sum(s[1] for s in stats)
if time_all > 0:
table = PrettyTable("#", "%", "time", "unit")
table.align["unit"] = "l"
top_time = 0
for i in range(1, min(top_number, len(stats)) + 1):
top_time += stats[i - 1][1]
table.add_row(i, int(stats[i - 1][1] * 100 / time_all),
datetime.timedelta(seconds=stats[i - 1][1]),
stats[i - 1][0])
table.add_row(u"Σ", int(top_time * 100 / time_all),
datetime.timedelta(seconds=top_time), "Top 5")
self.info(u"Unit run time statistics top:\n%s", table)
table = PrettyTable("units", "real", u"η,%")
table.add_row(datetime.timedelta(seconds=time_all),
datetime.timedelta(seconds=self._run_time_),
int(time_all * 100 / (self._run_time_ or 1)))
self.info(u"Total run time:\n%s", table)
table = PrettyTable("method", "%", "time")
table.align["method"] = "l"
time_all = 0
for k, v in sorted(self._method_time_.items()):
if k == "run":
continue
time_all += v
if self._run_time_ > 0:
table.add_row(k, int(v * 100 / self._run_time_),
datetime.timedelta(seconds=v))
if self.is_slave:
table.add_row(u"Σ", int(time_all * 100 / self._run_time_),
datetime.timedelta(seconds=time_all))
if time_all > 0:
self.info(u"Workflow methods run time:\n%s", table)
