def trainer(mgr, config, progress):
rebuild = config.training.rebuild
rescore = config.training.rescore
if not config.training.ham_folder_ids or not config.training.spam_folder_ids:
progress.error("You must specify at least one spam, and one good folder")
return
if rebuild:
# Make a new temporary bayes database to use for training.
# If we complete, then the manager "adopts" it.
# This prevents cancelled training from leaving a "bad" db, and
# also prevents mail coming in during training from being classified
# with the partial database.
import os, manager
bayes_base = os.path.join(mgr.data_directory, "$sbtemp$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory, "$sbtemp$default_message_database")
# determine which db manager to use, and create it.
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
else:
classifier_data = mgr.classifier_data
# We do this in possibly 3 stages - train, filter, save
# re-scoring is much slower and training (as we actually have to save
# the message back.)
# Saving is really slow sometimes, but we only have 1 tick for that anyway
if rescore:
stages = ("Training", .3), ("Saving", .1), ("Scoring", .6)
else:
stages = ("Training", .9), ("Saving", .1)
progress.set_stages(stages)
real_trainer(classifier_data, config, mgr.message_store, progress)
if progress.stop_requested():
return
if rebuild:
assert mgr.classifier_data is not classifier_data
mgr.classifier_data.Adopt(classifier_data)
classifier_data = mgr.classifier_data
progress.tick()
if rescore:
# Setup the "filter now" config to what we want.
config = mgr.config.filter_now
config.only_unread = False
config.only_unseen = False
config.action_all = False
config.folder_ids = mgr.config.training.ham_folder_ids + mgr.config.training.spam_folder_ids
config.include_sub = mgr.config.training.ham_include_sub or mgr.config.training.spam_include_sub
import filter
filter.filterer(mgr, mgr.config, progress)
bayes = classifier_data.bayes
progress.set_status("Completed training with %d spam and %d good messages" % (bayes.nspam, bayes.nham))
def WizardTrainer(mgr, config, progress):
import os, manager, train
bayes_base = os.path.join(mgr.data_directory,
"$sbwiz$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory,
"$sbwiz$default_message_database")
fnames = []
for ext in ".pck", ".db":
fnames.append(bayes_base + ext)
fnames.append(mdb_base + ext)
config.wizard.temp_training_names = fnames
# determine which db manager to use, and create it.
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
rescore = config.training.rescore
if rescore:
stages = (_("Training"), .3), (_("Saving"), .1), (_("Scoring"), .6)
else:
stages = (_("Training"), .9), (_("Saving"), .1)
progress.set_stages(stages)
train.real_trainer(classifier_data, config, mgr.message_store, progress)
# xxx - more hacks - we should pass the classifier data in.
orig_classifier_data = mgr.classifier_data
mgr.classifier_data = classifier_data # temporary
try:
progress.tick()
if rescore:
# Setup the "filter now" config to what we want.
now_config = config.filter_now
now_config.only_unread = False
now_config.only_unseen = False
now_config.action_all = False
now_config.folder_ids = config.training.ham_folder_ids + \
config.training.spam_folder_ids
now_config.include_sub = config.training.ham_include_sub or \
config.training.spam_include_sub
import filter
filter.filterer(mgr, config, progress)
bayes = classifier_data.bayes
progress.set_status(_("Completed training with %d spam and %d good messages") \
% (bayes.nspam, bayes.nham))
finally:
mgr.wizard_classifier_data = classifier_data
mgr.classifier_data = orig_classifier_data
def WizardTrainer(mgr, config, progress):
import os, manager, train
bayes_base = os.path.join(mgr.data_directory, "$sbwiz$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory, "$sbwiz$default_message_database")
fnames = []
for ext in ".pck", ".db":
fnames.append(bayes_base+ext)
fnames.append(mdb_base+ext)
config.wizard.temp_training_names = fnames
# determine which db manager to use, and create it.
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
rescore = config.training.rescore
if rescore:
stages = ("Training", .3), ("Saving", .1), ("Scoring", .6)
else:
stages = ("Training", .9), ("Saving", .1)
progress.set_stages(stages)
train.real_trainer(classifier_data, config, mgr.message_store, progress)
# xxx - more hacks - we should pass the classifier data in.
orig_classifier_data = mgr.classifier_data
mgr.classifier_data = classifier_data # temporary
try:
progress.tick()
if rescore:
# Setup the "filter now" config to what we want.
now_config = config.filter_now
now_config.only_unread = False
now_config.only_unseen = False
now_config.action_all = False
now_config.folder_ids = config.training.ham_folder_ids + \
config.training.spam_folder_ids
now_config.include_sub = config.training.ham_include_sub or \
config.training.spam_include_sub
import filter
filter.filterer(mgr, config, progress)
bayes = classifier_data.bayes
progress.set_status("Completed training with %d spam and %d good messages" \
% (bayes.nspam, bayes.nham))
finally:
mgr.wizard_classifier_data = classifier_data
mgr.classifier_data = orig_classifier_data
def trainer(mgr, config, progress):
rebuild = config.training.rebuild
rescore = config.training.rescore
if not config.training.ham_folder_ids and not config.training.spam_folder_ids:
progress.error(_("You must specify at least one spam or one good folder"))
return
if rebuild:
import os, manager
bayes_base = os.path.join(mgr.data_directory, "$sbtemp$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory, "$sbtemp$default_message_database")
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
else:
classifier_data = mgr.classifier_data
if rescore:
stages = (_("Training"), 0.3), (_("Saving"), 0.1), (_("Scoring"), 0.6)
else:
stages = (_("Training"), 0.9), (_("Saving"), 0.1)
progress.set_stages(stages)
real_trainer(classifier_data, config, mgr.message_store, progress)
if progress.stop_requested():
return
if rebuild:
assert mgr.classifier_data is not classifier_data
mgr.AdoptClassifierData(classifier_data)
classifier_data = mgr.classifier_data
mgr.LogDebug(1, "Session:" + "\r\n".join(mgr.stats.GetStats(session_only=True)))
mgr.LogDebug(1, "Total:" + "\r\n".join(mgr.stats.GetStats()))
mgr.stats.Reset()
mgr.stats.ResetTotal(permanently=True)
progress.tick()
if rescore:
config = mgr.config.filter_now
config.only_unread = False
config.only_unseen = False
config.action_all = False
config.folder_ids = mgr.config.training.ham_folder_ids + mgr.config.training.spam_folder_ids
config.include_sub = mgr.config.training.ham_include_sub or mgr.config.training.spam_include_sub
import filter
filter.filterer(mgr, mgr.config, progress)
bayes = classifier_data.bayes
progress.set_status(_("Completed training with %d spam and %d good messages") % (bayes.nspam, bayes.nham))
def WizardTrainer(mgr, config, progress):
import os, manager, train
bayes_base = os.path.join(mgr.data_directory, "$sbwiz$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory, "$sbwiz$default_message_database")
fnames = []
for ext in ".pck", ".db":
fnames.append(bayes_base + ext)
fnames.append(mdb_base + ext)
config.wizard.temp_training_names = fnames
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
rescore = config.training.rescore
if rescore:
stages = (_("Training"), 0.3), (_("Saving"), 0.1), (_("Scoring"), 0.6)
else:
stages = (_("Training"), 0.9), (_("Saving"), 0.1)
progress.set_stages(stages)
train.real_trainer(classifier_data, config, mgr.message_store, progress)
orig_classifier_data = mgr.classifier_data
mgr.classifier_data = classifier_data # temporary
try:
progress.tick()
if rescore:
now_config = config.filter_now
now_config.only_unread = False
now_config.only_unseen = False
now_config.action_all = False
now_config.folder_ids = config.training.ham_folder_ids + config.training.spam_folder_ids
now_config.include_sub = config.training.ham_include_sub or config.training.spam_include_sub
import filter
filter.filterer(mgr, config, progress)
bayes = classifier_data.bayes
progress.set_status(_("Completed training with %d spam and %d good messages") % (bayes.nspam, bayes.nham))
finally:
mgr.wizard_classifier_data = classifier_data
mgr.classifier_data = orig_classifier_data
def trainer(mgr, config, progress):
rebuild = config.training.rebuild
rescore = config.training.rescore
if not config.training.ham_folder_ids and not config.training.spam_folder_ids:
progress.error(
_("You must specify at least one spam or one good folder"))
return
if rebuild:
# Make a new temporary bayes database to use for training.
# If we complete, then the manager "adopts" it.
# This prevents cancelled training from leaving a "bad" db, and
# also prevents mail coming in during training from being classified
# with the partial database.
import os, manager
bayes_base = os.path.join(mgr.data_directory,
"$sbtemp$default_bayes_database")
mdb_base = os.path.join(mgr.data_directory,
"$sbtemp$default_message_database")
# determine which db manager to use, and create it.
ManagerClass = manager.GetStorageManagerClass()
db_manager = ManagerClass(bayes_base, mdb_base)
classifier_data = manager.ClassifierData(db_manager, mgr)
classifier_data.InitNew()
else:
classifier_data = mgr.classifier_data
# We do this in possibly 3 stages - train, filter, save
# re-scoring is much slower than training (as we actually have to save
# the message back.)
# Saving is really slow sometimes, but we only have 1 tick for that anyway
if rescore:
stages = (_("Training"), .3), (_("Saving"), .1), (_("Scoring"), .6)
else:
stages = (_("Training"), .9), (_("Saving"), .1)
progress.set_stages(stages)
real_trainer(classifier_data, config, mgr.message_store, progress)
if progress.stop_requested():
return
if rebuild:
assert mgr.classifier_data is not classifier_data
mgr.AdoptClassifierData(classifier_data)
classifier_data = mgr.classifier_data
# If we are rebuilding, then we reset the statistics, too.
# (But output them to the log for reference).
mgr.LogDebug(1, "Session:" + "\r\n".join(\
mgr.stats.GetStats(session_only=True)))
mgr.LogDebug(1, "Total:" + "\r\n".join(mgr.stats.GetStats()))
mgr.stats.Reset()
mgr.stats.ResetTotal(permanently=True)
progress.tick()
if rescore:
# Setup the "filter now" config to what we want.
config = mgr.config.filter_now
config.only_unread = False
config.only_unseen = False
config.action_all = False
config.folder_ids = mgr.config.training.ham_folder_ids + mgr.config.training.spam_folder_ids
config.include_sub = mgr.config.training.ham_include_sub or mgr.config.training.spam_include_sub
import filter
filter.filterer(mgr, mgr.config, progress)
bayes = classifier_data.bayes
progress.set_status(
_("Completed training with %d spam and %d good messages") %
(bayes.nspam, bayes.nham))