def load_classifier(self, check=True, quiet=False):
clfdir = self._settings('Classification',
'%s_classification_envpath' % self._channel)
if not isdir(clfdir):
return
else:
chid = self._settings('ObjectDetection',
'%s_channelid' % self._channel)
title = self._settings(
'Classification',
'%s_classification_regionname' % self._channel)
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels = {self._channel.title(): title},
color_channel=chid)
state = self._learner.state
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % state['path_env']
msg += 'Found class definition: %s\n' % b(
state['has_definition'])
msg += 'Found annotations: %s\n' % b(
state['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(
self._learner.is_annotated)
msg += 'Found ARFF file: %s\n' % b(state['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(
self._learner.is_trained)
msg += 'Found SVM model: %s\n' % b(state['has_model'])
msg += 'Found SVM range: %s\n' % b(state['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' \
%b(self._learner.is_valid)
msg += 'Found samples: %s\n' % b(state['has_path_samples'])
msg += (
'Sample images are only used for visualization and annotation '
' control at the moment.')
txt = '%s classifier inspection states' % self._channel
if not quiet:
information(self, txt, info=msg)
if state['has_arff']:
self._learner.importFromArff()
if state['has_definition']:
self._learner.loadDefinition()
def setup_classifiers(self):
sttg = self.settings
# processing channel, color channel
for p_channel, c_channel in self.ch_mapping.iteritems():
self.settings.set_section('Processing')
if sttg.get2(self._resolve_name(p_channel, 'classification')):
sttg.set_section('Classification')
clf = CommonClassPredictor(
clf_dir=sttg.get2(self._resolve_name(p_channel,
'classification_envpath')),
name=p_channel,
channels=self._channel_regions(p_channel),
color_channel=c_channel)
clf.importFromArff()
clf.loadClassifier()
self.classifiers[p_channel] = clf
def load_classifier(self, check=True):
_resolve = lambda x,y: self._settings.get(x, '%s_%s' % (self._channel, y))
env_path = convert_package_path(_resolve('Classification',
'classification_envpath'))
classifier_infos = {'strEnvPath' : env_path,
#'strModelPrefix' : _resolve('Classification', 'classification_prefix'),
'strChannelId' : _resolve('ObjectDetection', 'channelid'),
'strRegionId' : _resolve('Classification', 'classification_regionname'),
}
try:
self._learner = CommonClassPredictor(dctCollectSamples=classifier_infos)
except:
exception(self, 'Error on loading classifier.')
else:
result = self._learner.check()
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % result['path_env']
msg += 'Found class definition: %s\n' % b(result['has_definition'])
msg += 'Found annotations: %s\n' % b(result['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(self.is_pick_samples())
msg += 'Found ARFF file: %s\n' % b(result['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(self.is_train_classifier())
msg += 'Found SVM model: %s\n' % b(result['has_model'])
msg += 'Found SVM range: %s\n' % b(result['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' % b(self.is_apply_classifier())
msg += 'Found samples: %s\n' % b(result['has_path_samples'])
msg += 'Sample images are only used for visualization and annotation control at the moment.'
txt = '%s classifier inspection results' % self._channel
information(self, txt, info=msg)
if result['has_arff']:
self._learner.importFromArff()
nr_features_prev = len(self._learner.lstFeatureNames)
removed_features = self._learner.filterData(apply=False)
nr_features = nr_features_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES % (nr_features, nr_features_prev))
self._label_features.setToolTip("removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
if result['has_definition']:
self._learner.loadDefinition()
if result['has_conf']:
c, g, conf = self._learner.importConfusion()
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
else:
conf = None
self._init_conf_table(conf)
self._set_info_table(conf)
def load_classifier(self, check=True):
_resolve = lambda x,y: self._settings.get(x, '%s_%s'
% (self._channel, y))
clfdir = CecogEnvironment.convert_package_path(_resolve('Classification',
'classification_envpath'))
# XXX - where does the "." come
if not isdir(clfdir) or clfdir == ".":
return
else:
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels={self._channel.title(): _resolve('Classification', 'classification_regionname')},
color_channel=_resolve('ObjectDetection', 'channelid'))
result = self._learner.check()
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % result['path_env']
msg += 'Found class definition: %s\n' % b(result['has_definition'])
msg += 'Found annotations: %s\n' % b(result['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(self.is_pick_samples())
msg += 'Found ARFF file: %s\n' % b(result['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(self.is_train_classifier())
msg += 'Found SVM model: %s\n' % b(result['has_model'])
msg += 'Found SVM range: %s\n' % b(result['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' % b(self.is_apply_classifier())
msg += 'Found samples: %s\n' % b(result['has_path_samples'])
msg += 'Sample images are only used for visualization and annotation control at the moment.'
txt = '%s classifier inspection results' % self._channel
information(self, txt, info=msg)
if result['has_arff']:
self._learner.importFromArff()
nr_features_prev = len(self._learner.feature_names)
removed_features = self._learner.filter_nans(apply=True)
nr_features = nr_features_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES %(nr_features, nr_features_prev))
self._label_features.setToolTip("removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
if result['has_definition']:
self._learner.loadDefinition()
if result['has_conf']:
c, g, conf = self._learner.importConfusion()
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
else:
conf = None
self._init_conf_table(conf)
self._set_info_table(conf)
def setup_classifiers(self):
sttg = self.settings
# processing channel, color channel
for p_channel, c_channel in self.ch_mapping.iteritems():
self.settings.set_section('Processing')
if sttg.get2(self._resolve_name(p_channel, 'classification')):
chreg = self._channel_regions(p_channel)
if sttg("EventSelection", "unsupervised_event_selection"):
nclusters = sttg("EventSelection", "num_clusters")
self.classifiers[p_channel] = ClassDefinitionUnsup( \
nclusters, chreg)
else:
sttg.set_section('Classification')
clf = CommonClassPredictor(
clf_dir=sttg.get2(self._resolve_name(p_channel,
'classification_envpath')),
name=p_channel,
channels=chreg,
color_channel=c_channel)
clf.importFromArff()
clf.loadClassifier()
self.classifiers[p_channel] = clf
def _setupClassifier(self):
for name, image in self.images.iteritems():
if image is None:
continue
self.classifiers[name] = CommonClassPredictor( \
clf_dir=self.mapper('Classification',
'%s_classification_envpath'
%name),
name=name,
channels=name,
color_channel=self.mapper("ObjectDetection", "%s_channelid" %name))
self.classifiers[name].importFromArff()
self.classifiers[name].loadClassifier()
def load_classifier(self, check=True, quiet=False):
clfdir = self._settings('Classification',
'%s_classification_envpath' %self._channel)
if not isdir(clfdir):
return
else:
chid = self._settings('ObjectDetection', '%s_channelid' %self._channel)
title = self._settings('Classification',
'%s_classification_regionname' %self._channel)
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels = {self._channel.title(): title},
color_channel=chid)
state = self._learner.state
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % state['path_env']
msg += 'Found class definition: %s\n' % b(state['has_definition'])
msg += 'Found annotations: %s\n' % b(state['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(self._learner.is_annotated)
msg += 'Found ARFF file: %s\n' % b(state['has_arff'])
msg += 'Can you train a classifier? %s\n\n' %b(self._learner.is_trained)
msg += 'Found SVM model: %s\n' % b(state['has_model'])
msg += 'Found SVM range: %s\n' % b(state['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' \
%b(self._learner.is_valid)
msg += 'Found samples: %s\n' % b(state['has_path_samples'])
msg += ('Sample images are only used for visualization and annotation '
' control at the moment.')
txt = '%s classifier inspection states' % self._channel
if not quiet:
information(self, txt, info=msg)
if state['has_arff']:
self._learner.importFromArff()
if state['has_definition']:
self._learner.loadDefinition()
def setup_classifiers(self):
sttg = self.settings
# processing channel, color channel
for p_channel, c_channel in self.ch_mapping.iteritems():
self.settings.set_section('Processing')
if sttg.get2(self._resolve_name(p_channel, 'classification')):
sttg.set_section('Classification')
clf = CommonClassPredictor(
clf_dir=sttg.get2(
self._resolve_name(p_channel,
'classification_envpath')),
name=p_channel,
channels=self._channel_regions(p_channel),
color_channel=c_channel)
clf.importFromArff()
clf.loadClassifier()
self.classifiers[p_channel] = clf
def setup_classifiers(self):
sttg = self.settings
# processing channel, color channel
for p_channel, c_channel in self.ch_mapping.iteritems():
self.settings.set_section('Processing')
if sttg.get2(self._resolve_name(p_channel, 'classification')):
chreg = self._channel_regions(p_channel)
if sttg("EventSelection", "unsupervised_event_selection"):
nclusters = sttg("EventSelection", "num_clusters")
self.classifiers[p_channel] = ClassDefinitionUnsup( \
nclusters, chreg)
else:
sttg.set_section('Classification')
clf = CommonClassPredictor(
clf_dir=sttg.get2(self._resolve_name(p_channel,
'classification_envpath')),
name=p_channel,
channels=chreg,
color_channel=c_channel)
clf.importFromArff()
clf.loadClassifier()
self.classifiers[p_channel] = clf
def _on_process_start(self, name, start_again=False):
if not self._is_running or start_again:
is_valid = True
self._is_abort = False
self._has_error = False
if self._process_items is None:
cls = self._control_buttons[name]['cls']
if type(cls) == types.ListType:
self._process_items = cls
self._current_process_item = 0
cls = cls[self._current_process_item]
else:
self._process_items = None
self._current_process_item = 0
else:
cls = self._process_items[self._current_process_item]
if self.name == SECTION_NAME_CLASSIFICATION:
result_frame = self._get_result_frame(self._tab_name)
result_frame.load_classifier(check=False)
learner = result_frame._learner
if name == self.PROCESS_PICKING:
if not result_frame.classifier.is_annotated:
is_valid = False
result_frame.msg_pick_samples(self)
elif result_frame.classifier.is_trained:
if not question(self, 'Samples already picked',
'Do you want to pick samples again and '
'overwrite previous '
'results?'):
is_valid = False
elif name == self.PROCESS_TRAINING:
if not result_frame.classifier.is_trained:
is_valid = False
result_frame.msg_train_classifier(self)
elif result_frame.classifier.is_valid:
if not question(self, 'Classifier already trained',
'Do you want to train the classifier '
'again?'):
is_valid = False
elif name == self.PROCESS_TESTING and not result_frame.classifier.is_valid:
is_valid = False
result_frame.msg_apply_classifier(self)
if cls is MultiAnalyzerThread:
ncpu = cpu_count()
(ncpu, ok) = QInputDialog.getInt(None, "On your machine are %d processers available." % ncpu, \
"Select the number of processors", \
ncpu, 1, ncpu*2)
if not ok:
self._process_items = None
is_valid = False
if is_valid:
self._current_process = name
if not start_again:
self.parent().main_window.log_window.clear()
self._is_running = True
self._stage_infos = {}
self._toggle_tabs(False)
# disable all section button of the main widget
self.toggle_tabs.emit(self.get_name())
self._set_control_button_text(idx=1)
self._toggle_control_buttons()
imagecontainer = self.parent().main_window._imagecontainer
if cls is PickerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._clear_image()
elif cls is AnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._clear_image()
elif cls is TrainingThread:
self._current_settings = self._settings.copy()
self._analyzer = cls(self, self._current_settings, result_frame._learner)
self._analyzer.setTerminationEnabled(True)
self._analyzer.conf_result.connect(result_frame.on_conf_result,
Qt.QueuedConnection)
result_frame.reset()
elif cls is MultiAnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer, ncpu)
elif cls is ErrorCorrectionThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
self.parent().main_window._imagecontainer)
elif cls is PostProcessingThread:
learner_dict = {}
for channel in ['primary', 'secondary']:
path = self._settings('Classification', '%s_classification_envpath' %channel)
if (self._settings('Processing', '%s_classification' %channel) and
(channel == 'primary' or self._settings('General', 'process_secondary'))):
learner = CommonClassPredictor( \
path,
self._settings('ObjectDetection', '%s_channelid' %channel),
self._settings('Classification', '%s_classification_regionname' %channel))
learner.importFromArff()
learner_dict[channel] = learner
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, learner_dict, imagecontainer)
self._analyzer.setTerminationEnabled(True)
self._analyzer.finished.connect(self._on_process_finished)
self._analyzer.stage_info.connect(self._on_update_stage_info, Qt.QueuedConnection)
self._analyzer.analyzer_error.connect(self._on_error, Qt.QueuedConnection)
self._analyzer.image_ready.connect(self._on_update_image)
self._analyzer.start(QThread.LowestPriority)
if self._current_process_item == 0:
self.status_message.emit('Process started...')
else:
self._abort_processing()
def _on_process_start(self, name, start_again=False):
if not self._is_running or start_again:
is_valid = True
self._is_abort = False
self._has_error = False
if self._process_items is None:
cls = self._control_buttons[name]['cls']
if type(cls) == types.ListType:
self._process_items = cls
self._current_process_item = 0
cls = cls[self._current_process_item]
else:
self._process_items = None
self._current_process_item = 0
else:
cls = self._process_items[self._current_process_item]
if self.name == SECTION_NAME_CLASSIFICATION:
result_frame = self._get_result_frame(self._tab_name)
result_frame.load_classifier(check=False)
learner = result_frame._learner
if name == self.PROCESS_PICKING:
if not result_frame.classifier.is_annotated:
is_valid = False
result_frame.msg_pick_samples(self)
elif result_frame.classifier.is_trained:
if not question(self, 'Samples already picked',
'Do you want to pick samples again and '
'overwrite previous '
'results?'):
is_valid = False
elif name == self.PROCESS_TRAINING:
if not result_frame.classifier.is_trained:
is_valid = False
result_frame.msg_train_classifier(self)
elif result_frame.classifier.is_valid:
if not question(self, 'Classifier already trained',
'Do you want to train the classifier '
'again?'):
is_valid = False
elif name == self.PROCESS_TESTING and not result_frame.classifier.is_valid:
is_valid = False
result_frame.msg_apply_classifier(self)
if cls is MultiAnalyzerThread:
ncpu = cpu_count()
(ncpu, ok) = QInputDialog.getInt(None, "On your machine are %d processers available." % ncpu, \
"Select the number of processors", \
ncpu, 1, ncpu*2)
if not ok:
self._process_items = None
is_valid = False
if is_valid:
self._current_process = name
if not start_again:
self.parent().main_window.log_window.clear()
self._is_running = True
self._stage_infos = {}
self._toggle_tabs(False)
# disable all section button of the main widget
self.toggle_tabs.emit(self.get_name())
self._set_control_button_text(idx=1)
self._toggle_control_buttons()
imagecontainer = self.parent().main_window._imagecontainer
if cls is PickerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._clear_image()
elif cls is AnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._clear_image()
elif cls is TrainingThread:
self._current_settings = self._settings.copy()
self._analyzer = cls(self, self._current_settings, result_frame._learner)
self._analyzer.setTerminationEnabled(True)
self._analyzer.conf_result.connect(result_frame.on_conf_result,
Qt.QueuedConnection)
result_frame.reset()
elif cls is MultiAnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer, ncpu)
elif cls is ErrorCorrectionThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
self.parent().main_window._imagecontainer)
elif cls is PostProcessingThread:
learner_dict = {}
for channel in ['primary', 'secondary']:
path = self._settings('Classification', '%s_classification_envpath' %channel)
if (self._settings('Processing', '%s_classification' %channel) and
(channel == 'primary' or self._settings('General', 'process_secondary'))):
learner = CommonClassPredictor( \
path,
self._settings('ObjectDetection', '%s_channelid' %channel),
self._settings('Classification', '%s_classification_regionname' %channel))
learner.importFromArff()
learner_dict[channel] = learner
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, learner_dict, imagecontainer)
self._analyzer.setTerminationEnabled(True)
self._analyzer.finished.connect(self._on_process_finished)
self._analyzer.stage_info.connect(self._on_update_stage_info, Qt.QueuedConnection)
self._analyzer.analyzer_error.connect(self._on_error, Qt.QueuedConnection)
self._analyzer.image_ready.connect(self._on_update_image)
self._analyzer.start(QThread.LowestPriority)
if self._current_process_item == 0:
self.status_message.emit('Process started...')
else:
self._abort_processing()
class ClassifierResultFrame(QGroupBox):
LABEL_FEATURES = '#Features: %d (%d)'
LABEL_ACC = 'Overall accuracy: %.1f%%'
LABEL_C = 'Log2(C) = %.1f'
LABEL_G = 'Log2(g) = %.1f'
def __init__(self, parent, channel, settings):
super(ClassifierResultFrame, self).__init__(parent)
self._channel = channel
self._settings = settings
layout = QVBoxLayout(self)
layout.setContentsMargins(5, 5, 5, 5)
splitter = QSplitter(Qt.Horizontal, self)
splitter.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
splitter.setStretchFactor(0, 2)
layout.addWidget(splitter)
frame_info = QFrame()
layout_info = QVBoxLayout(frame_info)
label = QLabel('Class & annotation info', frame_info)
layout_info.addWidget(label)
self._table_info = QTableWidget(frame_info)
self._table_info.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_info.setSelectionMode(QTableWidget.NoSelection)
self._table_info.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_info.addWidget(self._table_info)
splitter.addWidget(frame_info)
frame_conf = QFrame()
layout_conf = QVBoxLayout(frame_conf)
label = QLabel('Confusion matrix', frame_conf)
layout_conf.addWidget(label)
self._table_conf = QTableWidget(frame_conf)
self._table_conf.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_conf.setSelectionMode(QTableWidget.NoSelection)
self._table_conf.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_conf.addWidget(self._table_conf)
splitter.addWidget(frame_conf)
desc = QFrame(self)
layout_desc = QHBoxLayout(desc)
self._label_acc = QLabel(self.LABEL_ACC % float('NAN'), desc)
layout_desc.addWidget(self._label_acc, Qt.AlignLeft)
self._label_features = QLabel(self.LABEL_FEATURES % (0,0), desc)
layout_desc.addWidget(self._label_features, Qt.AlignLeft)
self._label_c = QLabel(self.LABEL_C % float('NAN'), desc)
layout_desc.addWidget(self._label_c, Qt.AlignLeft)
self._label_g = QLabel(self.LABEL_G % float('NAN'), desc)
layout_desc.addWidget(self._label_g, Qt.AlignLeft)
self.browserBtn = QPushButton('Open browser for annotation', desc)
layout_desc.addWidget(self.browserBtn)
layout.addWidget(desc)
self._has_data = False
self._learner = None
@property
def classifier(self):
return self._learner
def clear(self):
self._table_conf.clear()
self._table_info.clear()
self._has_data = False
def reset(self):
self._has_data = False
self._table_conf.clearContents()
def on_load(self):
self.load_classifier(check=True)
self.update_frame()
def load_classifier(self, check=True, quiet=False):
clfdir = self._settings('Classification',
'%s_classification_envpath' %self._channel)
if not isdir(clfdir):
return
else:
chid = self._settings('ObjectDetection', '%s_channelid' %self._channel)
title = self._settings('Classification',
'%s_classification_regionname' %self._channel)
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels = {self._channel.title(): title},
color_channel=chid)
state = self._learner.state
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % state['path_env']
msg += 'Found class definition: %s\n' % b(state['has_definition'])
msg += 'Found annotations: %s\n' % b(state['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(self._learner.is_annotated)
msg += 'Found ARFF file: %s\n' % b(state['has_arff'])
msg += 'Can you train a classifier? %s\n\n' %b(self._learner.is_trained)
msg += 'Found SVM model: %s\n' % b(state['has_model'])
msg += 'Found SVM range: %s\n' % b(state['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' \
%b(self._learner.is_valid)
msg += 'Found samples: %s\n' % b(state['has_path_samples'])
msg += ('Sample images are only used for visualization and annotation '
' control at the moment.')
txt = '%s classifier inspection states' % self._channel
if not quiet:
information(self, txt, info=msg)
if state['has_arff']:
self._learner.importFromArff()
if state['has_definition']:
self._learner.loadDefinition()
def update_frame(self):
"""Updates cass & annotation info and confusion matrix in the gui"""
# update only if possible...
# if samples were picked/annotated
try:
nftr_prev = len(self._learner.feature_names)
except TypeError:
pass
except AttributeError: # ClassDefUnsupervised, nothing to draw
return
else:
removed_features = self._learner.filter_nans(apply=True)
nftr = nftr_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES %(nftr, nftr_prev))
self._label_features.setToolTip(
"removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
# if classifier was trained
try:
c, g, conf = self._learner.importConfusion()
except IOError as e:
conf = None
self._init_conf_table(conf)
else:
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
self._set_info_table(conf)
def msg_pick_samples(self, parent):
state = self._learner.state
text = 'Sample picking is not possible'
info = 'You need to provide a class definition '\
'file and annotation files.'
detail = 'Missing components:\n'
if not state['has_path_annotations']:
detail += "- Annotation path '%s' not found.\n" %state['path_annotations']
if not state['has_definition']:
detail += "- Class definition file '%s' not found.\n" %state['definition']
return information(parent, text, info, detail)
def msg_train_classifier(self, parent):
state = self._learner.state
text = 'Classifier training is not possible'
info = 'You need to pick samples first.'
detail = 'Missing components:\n'
if not state['has_arff']:
detail += "- Feature file '%s' not found.\n" % state['arff']
return information(parent, text, info, detail)
def msg_apply_classifier(self, parent):
state = self._learner.state
text = 'Classifier model not found'
info = 'You need to train a classifier first.'
detail = 'Missing components:\n'
if not state['has_model']:
detail += "- SVM model file '%s' not found.\n" % state['model']
if not state['has_range']:
detail += "- SVM range file '%s' not found.\n" % state['range']
return information(parent, text, info, detail)
def _set_info_table(self, conf):
rows = len(self._learner.class_labels)
self._table_info.clear()
names_horizontal = [('Name', 'class name'),
('Samples', 'class samples'),
('Color', 'class color'),
('%PR', 'class precision in %'),
('%SE', 'class sensitivity in %')]
names_vertical = [str(k) for k in self._learner.class_names.keys()] + \
['', '#']
self._table_info.setColumnCount(len(names_horizontal))
self._table_info.setRowCount(len(names_vertical))
self._table_info.setVerticalHeaderLabels(names_vertical)
self._table_info.setColumnWidth(1, 20)
for c, (name, info) in enumerate(names_horizontal):
item = QTableWidgetItem(name)
item.setToolTip(info)
self._table_info.setHorizontalHeaderItem(c, item)
for r, label in enumerate(self._learner.class_names.keys()):
self._table_info.setRowHeight(r, 20)
name = self._learner.class_names[label]
samples = self._learner.names2samples[name]
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(\
QColor(*hex2rgb(self._learner.hexcolors[name]))))
self._table_info.setItem(r, 2, item)
if not conf is None and r < len(conf):
item = QTableWidgetItem('%.1f' % (conf.ppv[r] * 100.))
item.setToolTip('"%s" precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.se[r] * 100.))
item.setToolTip('"%s" sensitivity' % name)
self._table_info.setItem(r, 4, item)
if not conf is None:
self._table_info.setRowHeight(r+1, 20)
r += 2
self._table_info.setRowHeight(r, 20)
name = "overal"
samples = sum(self._learner.names2samples.values())
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(QColor(*hex2rgb('#FFFFFF'))))
self._table_info.setItem(r, 2, item)
item = QTableWidgetItem('%.1f' % (conf.wav_ppv * 100.))
item.setToolTip('%s per class precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.wav_se * 100.))
item.setToolTip('%s per class sensitivity' % name)
self._table_info.setItem(r, 4, item)
self._table_info.resizeColumnsToContents()
def _init_conf_table(self, conf):
self._table_conf.clear()
if not conf is None:
conf_array = conf.conf
rows, cols = conf_array.shape
self._table_conf.setColumnCount(cols)
self._table_conf.setRowCount(rows)
for i, label in enumerate(self._learner.class_names):
h_item = QTableWidgetItem(str(label))
v_item = QTableWidgetItem(str(label))
tooltip = '%d : %s' %(label, self._learner.class_names[label])
v_item.setToolTip(tooltip)
h_item.setToolTip(tooltip)
self._table_conf.setHorizontalHeaderItem(i, h_item)
self._table_conf.setVerticalHeaderItem(i, v_item)
self._table_conf.setColumnWidth(i, 20)
self._table_conf.setRowHeight(i, 20)
def _update_conf_table(self, conf):
conf_array = conf.conf
rows, cols = conf_array.shape
conf_norm = conf_array.swapaxes(0,1) / numpy.array(numpy.sum(conf_array, 1), numpy.float)
conf_norm = conf_norm.swapaxes(0,1)
self._table_conf.clearContents()
for r in range(rows):
for c in range(cols):
item = QTableWidgetItem()
item.setToolTip('%d samples' % conf_array[r,c])
if not numpy.isnan(conf_norm[r,c]):
col = int(255 * (1 - conf_norm[r,c]))
item.setBackground(QBrush(QColor(col, col, col)))
self._table_conf.setItem(r, c, item)
def _set_info(self, c, g, conf):
self._label_acc.setText(self.LABEL_ACC % (conf.ac_sample*100.))
self._label_c.setText(self.LABEL_C % c)
self._label_g.setText(self.LABEL_G % g)
def on_conf_result(self, c, g, conf):
self._set_info(c, g, conf)
if not self._has_data:
self._has_data = True
self._init_conf_table(conf)
self._set_info_table(conf)
self._update_conf_table(conf)
class ClassifierResultFrame(QGroupBox):
LABEL_FEATURES = '#Features: %d (%d)'
LABEL_ACC = 'Overall accuracy: %.1f%%'
LABEL_C = 'Log2(C) = %.1f'
LABEL_G = 'Log2(g) = %.1f'
def __init__(self, parent, channel, settings):
super(ClassifierResultFrame, self).__init__(parent)
self._channel = channel
self._settings = settings
layout = QVBoxLayout(self)
layout.setContentsMargins(5, 5, 5, 5)
splitter = QSplitter(Qt.Horizontal, self)
splitter.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
splitter.setStretchFactor(0, 2)
layout.addWidget(splitter)
frame_info = QFrame()
layout_info = QVBoxLayout(frame_info)
label = QLabel('Class & annotation info', frame_info)
layout_info.addWidget(label)
self._table_info = QTableWidget(frame_info)
self._table_info.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_info.setSelectionMode(QTableWidget.NoSelection)
self._table_info.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_info.addWidget(self._table_info)
splitter.addWidget(frame_info)
frame_conf = QFrame()
layout_conf = QVBoxLayout(frame_conf)
label = QLabel('Confusion matrix', frame_conf)
layout_conf.addWidget(label)
self._table_conf = QTableWidget(frame_conf)
self._table_conf.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_conf.setSelectionMode(QTableWidget.NoSelection)
self._table_conf.setSizePolicy(\
QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_conf.addWidget(self._table_conf)
splitter.addWidget(frame_conf)
desc = QFrame(self)
layout_desc = QHBoxLayout(desc)
self._label_acc = QLabel(self.LABEL_ACC % float('NAN'), desc)
layout_desc.addWidget(self._label_acc, Qt.AlignLeft)
self._label_features = QLabel(self.LABEL_FEATURES % (0, 0), desc)
layout_desc.addWidget(self._label_features, Qt.AlignLeft)
self._label_c = QLabel(self.LABEL_C % float('NAN'), desc)
layout_desc.addWidget(self._label_c, Qt.AlignLeft)
self._label_g = QLabel(self.LABEL_G % float('NAN'), desc)
layout_desc.addWidget(self._label_g, Qt.AlignLeft)
self.browserBtn = QPushButton('Open browser for annotation', desc)
layout_desc.addWidget(self.browserBtn)
layout.addWidget(desc)
self._has_data = False
self._learner = None
@property
def classifier(self):
return self._learner
def clear(self):
self._table_conf.clear()
self._table_info.clear()
self._has_data = False
def reset(self):
self._has_data = False
self._table_conf.clearContents()
def on_load(self):
self.load_classifier(check=True)
self.update_frame()
def load_classifier(self, check=True, quiet=False):
clfdir = self._settings('Classification',
'%s_classification_envpath' % self._channel)
if not isdir(clfdir):
return
else:
chid = self._settings('ObjectDetection',
'%s_channelid' % self._channel)
title = self._settings(
'Classification',
'%s_classification_regionname' % self._channel)
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels = {self._channel.title(): title},
color_channel=chid)
state = self._learner.state
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % state['path_env']
msg += 'Found class definition: %s\n' % b(
state['has_definition'])
msg += 'Found annotations: %s\n' % b(
state['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(
self._learner.is_annotated)
msg += 'Found ARFF file: %s\n' % b(state['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(
self._learner.is_trained)
msg += 'Found SVM model: %s\n' % b(state['has_model'])
msg += 'Found SVM range: %s\n' % b(state['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' \
%b(self._learner.is_valid)
msg += 'Found samples: %s\n' % b(state['has_path_samples'])
msg += (
'Sample images are only used for visualization and annotation '
' control at the moment.')
txt = '%s classifier inspection states' % self._channel
if not quiet:
information(self, txt, info=msg)
if state['has_arff']:
self._learner.importFromArff()
if state['has_definition']:
self._learner.loadDefinition()
def update_frame(self):
"""Updates cass & annotation info and confusion matrix in the gui"""
# update only if possible...
# if samples were picked/annotated
try:
nftr_prev = len(self._learner.feature_names)
except TypeError:
pass
except AttributeError: # ClassDefUnsupervised, nothing to draw
return
else:
removed_features = self._learner.filter_nans(apply=True)
nftr = nftr_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES %
(nftr, nftr_prev))
self._label_features.setToolTip(
"removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
# if classifier was trained
try:
c, g, conf = self._learner.importConfusion()
except IOError as e:
conf = None
self._init_conf_table(conf)
else:
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
self._set_info_table(conf)
def msg_pick_samples(self, parent):
state = self._learner.state
text = 'Sample picking is not possible'
info = 'You need to provide a class definition '\
'file and annotation files.'
detail = 'Missing components:\n'
if not state['has_path_annotations']:
detail += "- Annotation path '%s' not found.\n" % state[
'path_annotations']
if not state['has_definition']:
detail += "- Class definition file '%s' not found.\n" % state[
'definition']
return information(parent, text, info, detail)
def msg_train_classifier(self, parent):
state = self._learner.state
text = 'Classifier training is not possible'
info = 'You need to pick samples first.'
detail = 'Missing components:\n'
if not state['has_arff']:
detail += "- Feature file '%s' not found.\n" % state['arff']
return information(parent, text, info, detail)
def msg_apply_classifier(self, parent):
state = self._learner.state
text = 'Classifier model not found'
info = 'You need to train a classifier first.'
detail = 'Missing components:\n'
if not state['has_model']:
detail += "- SVM model file '%s' not found.\n" % state['model']
if not state['has_range']:
detail += "- SVM range file '%s' not found.\n" % state['range']
return information(parent, text, info, detail)
def _set_info_table(self, conf):
rows = len(self._learner.class_labels)
self._table_info.clear()
names_horizontal = [('Name', 'class name'),
('Samples', 'class samples'),
('Color', 'class color'),
('%PR', 'class precision in %'),
('%SE', 'class sensitivity in %')]
names_vertical = [str(k) for k in self._learner.class_names.keys()] + \
['', '#']
self._table_info.setColumnCount(len(names_horizontal))
self._table_info.setRowCount(len(names_vertical))
self._table_info.setVerticalHeaderLabels(names_vertical)
self._table_info.setColumnWidth(1, 20)
for c, (name, info) in enumerate(names_horizontal):
item = QTableWidgetItem(name)
item.setToolTip(info)
self._table_info.setHorizontalHeaderItem(c, item)
for r, label in enumerate(self._learner.class_names.keys()):
self._table_info.setRowHeight(r, 20)
name = self._learner.class_names[label]
samples = self._learner.names2samples[name]
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(\
QColor(*hex2rgb(self._learner.hexcolors[name]))))
self._table_info.setItem(r, 2, item)
if not conf is None and r < len(conf):
item = QTableWidgetItem('%.1f' % (conf.ppv[r] * 100.))
item.setToolTip('"%s" precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.se[r] * 100.))
item.setToolTip('"%s" sensitivity' % name)
self._table_info.setItem(r, 4, item)
if not conf is None:
self._table_info.setRowHeight(r + 1, 20)
r += 2
self._table_info.setRowHeight(r, 20)
name = "overal"
samples = sum(self._learner.names2samples.values())
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(QColor(*hex2rgb('#FFFFFF'))))
self._table_info.setItem(r, 2, item)
item = QTableWidgetItem('%.1f' % (conf.wav_ppv * 100.))
item.setToolTip('%s per class precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.wav_se * 100.))
item.setToolTip('%s per class sensitivity' % name)
self._table_info.setItem(r, 4, item)
self._table_info.resizeColumnsToContents()
def _init_conf_table(self, conf):
self._table_conf.clear()
if not conf is None:
conf_array = conf.conf
rows, cols = conf_array.shape
self._table_conf.setColumnCount(cols)
self._table_conf.setRowCount(rows)
for i, label in enumerate(self._learner.class_names):
h_item = QTableWidgetItem(str(label))
v_item = QTableWidgetItem(str(label))
tooltip = '%d : %s' % (label, self._learner.class_names[label])
v_item.setToolTip(tooltip)
h_item.setToolTip(tooltip)
self._table_conf.setHorizontalHeaderItem(i, h_item)
self._table_conf.setVerticalHeaderItem(i, v_item)
self._table_conf.setColumnWidth(i, 20)
self._table_conf.setRowHeight(i, 20)
def _update_conf_table(self, conf):
conf_array = conf.conf
rows, cols = conf_array.shape
conf_norm = conf_array.swapaxes(0, 1) / numpy.array(
numpy.sum(conf_array, 1), numpy.float)
conf_norm = conf_norm.swapaxes(0, 1)
self._table_conf.clearContents()
for r in range(rows):
for c in range(cols):
item = QTableWidgetItem()
item.setToolTip('%d samples' % conf_array[r, c])
if not numpy.isnan(conf_norm[r, c]):
col = int(255 * (1 - conf_norm[r, c]))
item.setBackground(QBrush(QColor(col, col, col)))
self._table_conf.setItem(r, c, item)
def _set_info(self, c, g, conf):
self._label_acc.setText(self.LABEL_ACC % (conf.ac_sample * 100.))
self._label_c.setText(self.LABEL_C % c)
self._label_g.setText(self.LABEL_G % g)
def on_conf_result(self, c, g, conf):
self._set_info(c, g, conf)
if not self._has_data:
self._has_data = True
self._init_conf_table(conf)
self._set_info_table(conf)
self._update_conf_table(conf)
import os
import numpy
import argparse
from cecog.learning.learning import CommonClassPredictor
import time
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Process some integers.')
parser.add_argument('directory',
type=str,
help='Directory to the classifier')
args = parser.parse_args()
if os.path.isdir(args.directory):
learner = CommonClassPredictor(args.directory, None, None)
learner.importFromArff()
t0 = time.time()
n, c, g, conf = learner.gridSearch()
print "Grid search took: ", time.time() - t0
#c, g, conf = learner.importConfusion()
numpy.set_printoptions(linewidth=80)
print "Confusion Matrix:"
for row in conf.conf:
print row
else:
raise IOError("%s\n is not a valid directory" % args.directory)
#learner.statsFromConfusion(conf)
# #benchmark
class ClassifierResultFrame(QGroupBox):
LABEL_FEATURES = '#Features: %d (%d)'
LABEL_ACC = 'Overall accuracy: %.1f%%'
LABEL_C = 'Log2(C) = %.1f'
LABEL_G = 'Log2(g) = %.1f'
def __init__(self, parent, channel, settings):
QGroupBox.__init__(self, parent)
self._channel = channel
self._settings = settings
layout = QVBoxLayout(self)
layout.setContentsMargins(5, 5, 5, 5)
#self._button = QPushButton('Load', self)
#self.connect(self._button, SIGNAL('clicked()'), self._on_load)
#layout.addWidget(self._button, 1, 2)
splitter = QSplitter(Qt.Horizontal, self)
splitter.setSizePolicy(QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
splitter.setStretchFactor(0, 2)
layout.addWidget(splitter)
frame_info = QFrame()
layout_info = QVBoxLayout(frame_info)
label = QLabel('Class & annotation info', frame_info)
layout_info.addWidget(label)
self._table_info = QTableWidget(frame_info)
self._table_info.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_info.setSelectionMode(QTableWidget.NoSelection)
self._table_info.setSizePolicy(QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_info.addWidget(self._table_info)
splitter.addWidget(frame_info)
frame_conf = QFrame()
layout_conf = QVBoxLayout(frame_conf)
label = QLabel('Confusion matrix', frame_conf)
layout_conf.addWidget(label)
self._table_conf = QTableWidget(frame_conf)
self._table_conf.setEditTriggers(QTableWidget.NoEditTriggers)
self._table_conf.setSelectionMode(QTableWidget.NoSelection)
self._table_conf.setSizePolicy(QSizePolicy(QSizePolicy.Expanding|QSizePolicy.Maximum,
QSizePolicy.Expanding|QSizePolicy.Maximum))
layout_conf.addWidget(self._table_conf)
splitter.addWidget(frame_conf)
desc = QFrame(self)
layout_desc = QHBoxLayout(desc)
self._label_acc = QLabel(self.LABEL_ACC % float('NAN'), desc)
layout_desc.addWidget(self._label_acc, Qt.AlignLeft)
self._label_features = QLabel(self.LABEL_FEATURES % (0,0), desc)
layout_desc.addWidget(self._label_features, Qt.AlignLeft)
self._label_c = QLabel(self.LABEL_C % float('NAN'), desc)
layout_desc.addWidget(self._label_c, Qt.AlignLeft)
self._label_g = QLabel(self.LABEL_G % float('NAN'), desc)
layout_desc.addWidget(self._label_g, Qt.AlignLeft)
btn = QPushButton('Show Browser', desc)
btn.clicked.connect(qApp._main_window._on_browser_open)
layout_desc.addWidget(btn)
layout.addWidget(desc)
self._has_data = False
def clear(self):
self._table_conf.clear()
self._table_info.clear()
self._has_data = False
def reset(self):
self._has_data = False
self._table_conf.clearContents()
def on_load(self):
self.load_classifier(check=True)
def load_classifier(self, check=True):
_resolve = lambda x,y: self._settings.get(x, '%s_%s' % (self._channel, y))
env_path = convert_package_path(_resolve('Classification',
'classification_envpath'))
classifier_infos = {'strEnvPath' : env_path,
#'strModelPrefix' : _resolve('Classification', 'classification_prefix'),
'strChannelId' : _resolve('ObjectDetection', 'channelid'),
'strRegionId' : _resolve('Classification', 'classification_regionname'),
}
try:
self._learner = CommonClassPredictor(dctCollectSamples=classifier_infos)
except:
exception(self, 'Error on loading classifier.')
else:
result = self._learner.check()
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % result['path_env']
msg += 'Found class definition: %s\n' % b(result['has_definition'])
msg += 'Found annotations: %s\n' % b(result['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(self.is_pick_samples())
msg += 'Found ARFF file: %s\n' % b(result['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(self.is_train_classifier())
msg += 'Found SVM model: %s\n' % b(result['has_model'])
msg += 'Found SVM range: %s\n' % b(result['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' % b(self.is_apply_classifier())
msg += 'Found samples: %s\n' % b(result['has_path_samples'])
msg += 'Sample images are only used for visualization and annotation control at the moment.'
txt = '%s classifier inspection results' % self._channel
information(self, txt, info=msg)
if result['has_arff']:
self._learner.importFromArff()
nr_features_prev = len(self._learner.lstFeatureNames)
removed_features = self._learner.filterData(apply=False)
nr_features = nr_features_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES % (nr_features, nr_features_prev))
self._label_features.setToolTip("removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
if result['has_definition']:
self._learner.loadDefinition()
if result['has_conf']:
c, g, conf = self._learner.importConfusion()
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
else:
conf = None
self._init_conf_table(conf)
self._set_info_table(conf)
def msg_pick_samples(self, parent):
result = self._learner.check()
text = 'Sample picking is not possible'
info = 'You need to provide a class definition '\
'file and annotation files.'
detail = 'Missing components:\n'
if not result['has_path_annotations']:
detail += "- Annotation path '%s' not found.\n" % result['path_annotations']
if not result['has_definition']:
detail += "- Class definition file '%s' not found.\n" % result['definition']
return information(parent, text, info, detail)
def is_pick_samples(self):
result = self._learner.check()
return result['has_path_annotations'] and result['has_definition']
def msg_train_classifier(self, parent):
result = self._learner.check()
text = 'Classifier training is not possible'
info = 'You need to pick samples first.'
detail = 'Missing components:\n'
if not result['has_arff']:
detail += "- Feature file '%s' not found.\n" % result['arff']
return information(parent, text, info, detail)
def is_train_classifier(self):
result = self._learner.check()
return result['has_arff']
def msg_apply_classifier(self, parent):
result = self._learner.check()
text = 'Classifier model not found'
info = 'You need to train a classifier first.'
detail = 'Missing components:\n'
if not result['has_model']:
detail += "- SVM model file '%s' not found.\n" % result['model']
if not result['has_range']:
detail += "- SVM range file '%s' not found.\n" % result['range']
return information(parent, text, info, detail)
def is_apply_classifier(self):
result = self._learner.check()
return result['has_model'] and result['has_range']
def _set_info_table(self, conf):
rows = len(self._learner.lstClassLabels)
self._table_info.clear()
names_horizontal = [('Name', 'class name'),
('Samples', 'class samples'),
('Color', 'class color'),
('%PR', 'class precision in %'),
('%SE', 'class sensitivity in %'),
# ('AC%', 'class accuracy in %'),
# ('SE%', 'class sensitivity in %'),
# ('SP%', 'class specificity in %'),
# ('PPV%', 'class positive predictive value in %'),
# ('NPV%', 'class negative predictive value in %'),
]
names_vertical = [str(self._learner.nl2l[r]) for r in range(rows)] + ['','#']
self._table_info.setColumnCount(len(names_horizontal))
self._table_info.setRowCount(len(names_vertical))
self._table_info.setVerticalHeaderLabels(names_vertical)
self._table_info.setColumnWidth(1, 20)
for c, (name, info) in enumerate(names_horizontal):
item = QTableWidgetItem(name)
item.setToolTip(info)
self._table_info.setHorizontalHeaderItem(c, item)
r = 0
for r in range(rows):
self._table_info.setRowHeight(r, 20)
label = self._learner.nl2l[r]
name = self._learner.dctClassNames[label]
samples = self._learner.names2samples[name]
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(QColor(*hexToRgb(self._learner.dctHexColors[name]))))
self._table_info.setItem(r, 2, item)
if not conf is None and r < len(conf):
item = QTableWidgetItem('%.1f' % (conf.ppv[r] * 100.))
item.setToolTip('"%s" precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.se[r] * 100.))
item.setToolTip('"%s" sensitivity' % name)
self._table_info.setItem(r, 4, item)
# item = QTableWidgetItem('%.1f' % (conf.ac[r] * 100.))
# item.setToolTip('"%s" accuracy' % name)
# self._table_info.setItem(r, 3, item)
#
# item = QTableWidgetItem('%.1f' % (conf.se[r] * 100.))
# item.setToolTip('"%s" sensitivity' % name)
# self._table_info.setItem(r, 4, item)
#
# item = QTableWidgetItem('%.1f' % (conf.sp[r] * 100.))
# item.setToolTip('"%s" specificity' % name)
# self._table_info.setItem(r, 5, item)
#
# item = QTableWidgetItem('%.1f' % (conf.ppv[r] * 100.))
# item.setToolTip('"%s" positive predictive value' % name)
# self._table_info.setItem(r, 6, item)
#
# item = QTableWidgetItem('%.1f' % (conf.npv[r] * 100.))
# item.setToolTip('"%s" negative predictive value' % name)
# self._table_info.setItem(r, 7, item)
if not conf is None:
self._table_info.setRowHeight(r+1, 20)
r += 2
self._table_info.setRowHeight(r, 20)
name = "overal"
samples = sum(self._learner.names2samples.values())
self._table_info.setItem(r, 0, QTableWidgetItem(name))
self._table_info.setItem(r, 1, QTableWidgetItem(str(samples)))
item = QTableWidgetItem(' ')
item.setBackground(QBrush(QColor(*hexToRgb('#FFFFFF'))))
self._table_info.setItem(r, 2, item)
item = QTableWidgetItem('%.1f' % (conf.wav_ppv * 100.))
item.setToolTip('%s per class precision' % name)
self._table_info.setItem(r, 3, item)
item = QTableWidgetItem('%.1f' % (conf.wav_se * 100.))
item.setToolTip('%s per class sensitivity' % name)
self._table_info.setItem(r, 4, item)
# item = QTableWidgetItem('%.1f' % (conf.av_ac * 100.))
# item.setToolTip('%s per class accuracy' % name)
# self._table_info.setItem(r, 3, item)
#
# item = QTableWidgetItem('%.1f' % (conf.av_se * 100.))
# item.setToolTip('%s per class sensitivity' % name)
# self._table_info.setItem(r, 4, item)
#
# item = QTableWidgetItem('%.1f' % (conf.av_sp * 100.))
# item.setToolTip('%s per class specificity' % name)
# self._table_info.setItem(r, 5, item)
#
# item = QTableWidgetItem('%.1f' % (conf.av_ppv * 100.))
# item.setToolTip('%s per class positive predictive value' % name)
# self._table_info.setItem(r, 6, item)
#
# item = QTableWidgetItem('%.1f' % (conf.av_npv * 100.))
# item.setToolTip('%s per class negative predictive value' % name)
# self._table_info.setItem(r, 7, item)
self._table_info.resizeColumnsToContents()
def _init_conf_table(self, conf):
self._table_conf.clear()
if not conf is None:
conf_array = conf.conf
rows, cols = conf_array.shape
self._table_conf.setColumnCount(cols)
self._table_conf.setRowCount(rows)
#names2cols = self._learner.dctHexColors
for c in range(cols):
self._table_conf.setColumnWidth(c, 20)
label = self._learner.nl2l[c]
name = self._learner.dctClassNames[label]
item = QTableWidgetItem(str(label))
item.setToolTip('%d : %s' % (label, name))
#item.setBackground(QBrush(QColor(*hexToRgb(names2cols[name]))))
self._table_conf.setHorizontalHeaderItem(c, item)
for r in range(rows):
self._table_conf.setRowHeight(r, 20)
label = self._learner.nl2l[r]
name = self._learner.dctClassNames[label]
item = QTableWidgetItem(str(label))
item.setToolTip('%d : %s' % (label, name))
#item.setForeground(QBrush(QColor(*hexToRgb(names2cols[name]))))
self._table_conf.setVerticalHeaderItem(r, item)
def _update_conf_table(self, conf):
conf_array = conf.conf
rows, cols = conf_array.shape
conf_norm = conf_array.swapaxes(0,1) / numpy.array(numpy.sum(conf_array, 1), numpy.float)
conf_norm = conf_norm.swapaxes(0,1)
self._table_conf.clearContents()
for r in range(rows):
for c in range(cols):
item = QTableWidgetItem()
item.setToolTip('%d samples' % conf_array[r,c])
if not numpy.isnan(conf_norm[r,c]):
col = int(255 * (1 - conf_norm[r,c]))
item.setBackground(QBrush(QColor(col, col, col)))
self._table_conf.setItem(r, c, item)
def _set_info(self, c, g, conf):
self._label_acc.setText(self.LABEL_ACC % (conf.ac_sample*100.))
self._label_c.setText(self.LABEL_C % c)
self._label_g.setText(self.LABEL_G % g)
def on_conf_result(self, c, g, conf):
self._set_info(c, g, conf)
if not self._has_data:
self._has_data = True
self._init_conf_table(conf)
self._set_info_table(conf)
self._update_conf_table(conf)
def load_classifier(self, check=True):
_resolve = lambda x, y: self._settings.get(
x, '%s_%s' % (self._channel, y))
clfdir = CecogEnvironment.convert_package_path(
_resolve('Classification', 'classification_envpath'))
# XXX - where does the "." come
if not isdir(clfdir) or clfdir == ".":
return
else:
self._learner = CommonClassPredictor( \
clf_dir=clfdir,
name=self._channel,
channels={self._channel.title(): _resolve('Classification', 'classification_regionname')},
color_channel=_resolve('ObjectDetection', 'channelid'))
result = self._learner.check()
if check:
b = lambda x: 'Yes' if x else 'No'
msg = 'Classifier path: %s\n' % result['path_env']
msg += 'Found class definition: %s\n' % b(
result['has_definition'])
msg += 'Found annotations: %s\n' % b(
result['has_path_annotations'])
msg += 'Can you pick new samples? %s\n\n' % b(
self.is_pick_samples())
msg += 'Found ARFF file: %s\n' % b(result['has_arff'])
msg += 'Can you train a classifier? %s\n\n' % b(
self.is_train_classifier())
msg += 'Found SVM model: %s\n' % b(result['has_model'])
msg += 'Found SVM range: %s\n' % b(result['has_range'])
msg += 'Can you apply the classifier to images? %s\n\n' % b(
self.is_apply_classifier())
msg += 'Found samples: %s\n' % b(result['has_path_samples'])
msg += 'Sample images are only used for visualization and annotation control at the moment.'
txt = '%s classifier inspection results' % self._channel
information(self, txt, info=msg)
if result['has_arff']:
self._learner.importFromArff()
nr_features_prev = len(self._learner.feature_names)
removed_features = self._learner.filter_nans(apply=True)
nr_features = nr_features_prev - len(removed_features)
self._label_features.setText(self.LABEL_FEATURES %
(nr_features, nr_features_prev))
self._label_features.setToolTip(
"removed %d features containing NA values:\n%s" %
(len(removed_features), "\n".join(removed_features)))
if result['has_definition']:
self._learner.loadDefinition()
if result['has_conf']:
c, g, conf = self._learner.importConfusion()
self._set_info(c, g, conf)
self._init_conf_table(conf)
self._update_conf_table(conf)
else:
conf = None
self._init_conf_table(conf)
self._set_info_table(conf)
def _on_process_start(self, name, start_again=False):
if not self._is_running or start_again:
is_valid = True
self._is_abort = False
self._has_error = False
if self._process_items is None:
cls = self._control_buttons[name]['cls']
if type(cls) == types.ListType:
self._process_items = cls
self._current_process_item = 0
cls = cls[self._current_process_item]
# remove HmmThread if process is not first in list and
# not valid error correction was activated
if (HmmThread in self._process_items
and self._process_items.index(HmmThread) > 0 and
not (self._settings.get(
'Processing', 'primary_errorcorrection') or
(self._settings.get(
'Processing', 'secondary_errorcorrection')
and self._settings.get(
'Processing',
'secondary_processchannel')))):
self._process_items.remove(HmmThread)
else:
self._process_items = None
self._current_process_item = 0
else:
cls = self._process_items[self._current_process_item]
if self.SECTION_NAME == 'Classification':
result_frame = self._get_result_frame(self._tab_name)
result_frame.load_classifier(check=False)
learner = result_frame._learner
if name == self.PROCESS_PICKING:
if not result_frame.is_pick_samples():
is_valid = False
result_frame.msg_pick_samples(self)
elif result_frame.is_train_classifier():
if not question(
self, 'Samples already picked',
'Do you want to pick samples again and '
'overwrite previous '
'results?'):
is_valid = False
elif name == self.PROCESS_TRAINING:
if not result_frame.is_train_classifier():
is_valid = False
result_frame.msg_train_classifier(self)
elif result_frame.is_apply_classifier():
if not question(
self, 'Classifier already trained',
'Do you want to train the classifier '
'again?'):
is_valid = False
elif name == self.PROCESS_TESTING and not result_frame.is_apply_classifier(
):
is_valid = False
result_frame.msg_apply_classifier(self)
elif cls is HmmThread:
success, cmd = HmmThread.test_executable(
self._settings.get('ErrorCorrection', 'filename_to_R'))
if not success:
critical(self, 'Error running R',
"The R command line program '%s' could not be executed.\n\n"\
"Make sure that the R-project is installed.\n\n"\
"See README.txt for details." % cmd)
is_valid = False
elif cls is MultiAnalyzerThread:
ncpu = cpu_count()
(ncpu, ok) = QInputDialog.getInt(None, "On your machine are %d processers available." % ncpu, \
"Select the number of processors", \
ncpu, 1, ncpu*2)
if not ok:
self._process_items = None
is_valid = False
if is_valid:
self._current_process = name
if not start_again:
self.parent().main_window.log_window.clear()
self._is_running = True
self._stage_infos = {}
self._toggle_tabs(False)
# disable all section button of the main widget
self.toggle_tabs.emit(self.get_name())
self._set_control_button_text(idx=1)
self._toggle_control_buttons()
imagecontainer = self.parent().main_window._imagecontainer
if cls is PickerThread:
self._current_settings = self._get_modified_settings(
name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
imagecontainer)
self._set_display_renderer_info()
self._clear_image()
elif cls is AnalyzerThread:
self._current_settings = self._get_modified_settings(
name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
imagecontainer)
self._set_display_renderer_info()
self._clear_image()
elif cls is TrainingThread:
self._current_settings = self._settings.copy()
self._analyzer = cls(self, self._current_settings,
result_frame._learner)
self._analyzer.setTerminationEnabled(True)
self._analyzer.conf_result.connect(
result_frame.on_conf_result, Qt.QueuedConnection)
result_frame.reset()
elif cls is MultiAnalyzerThread:
self._current_settings = self._get_modified_settings(
name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
imagecontainer, ncpu)
self._set_display_renderer_info()
elif cls is HmmThread:
self._current_settings = self._get_modified_settings(
name, imagecontainer.has_timelapse)
# FIXME: classifier handling needs revision!!!
learner_dict = {}
for kind in ['primary', 'secondary']:
_resolve = lambda x, y: self._settings.get(
x, '%s_%s' % (kind, y))
env_path = CecogEnvironment.convert_package_path(
_resolve('Classification',
'classification_envpath'))
if (os.path.exists(env_path) and
(kind == 'primary' or self._settings.get(
'Processing', 'secondary_processchannel'))):
learner = CommonClassPredictor( \
env_path,
_resolve('ObjectDetection', 'channelid'),
_resolve('Classification', 'classification_regionname'))
learner.importFromArff()
learner_dict[kind] = learner
### Whee, I like it... "self.parent().main_window._imagecontainer" crazy, crazy, michael... :-)
self._analyzer = cls(
self, self._current_settings, learner_dict,
self.parent().main_window._imagecontainer)
self._analyzer.setTerminationEnabled(True)
lw = self.parent().main_window.log_window
lw.show()
elif cls is PostProcessingThread:
learner_dict = {}
for kind in ['primary', 'secondary']:
_resolve = lambda x, y: self._settings.get(
x, '%s_%s' % (kind, y))
env_path = CecogEnvironment.convert_package_path(
_resolve('Classification',
'classification_envpath'))
if (_resolve('Processing', 'classification') and
(kind == 'primary' or self._settings.get(
'Processing', 'secondary_processchannel'))):
learner = CommonClassPredictor( \
env_path,
_resolve('ObjectDetection', 'channelid'),
_resolve('Classification', 'classification_regionname'))
learner.importFromArff()
learner_dict[kind] = learner
self._current_settings = self._get_modified_settings(
name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings,
learner_dict, imagecontainer)
self._analyzer.setTerminationEnabled(True)
self._analyzer.finished.connect(self._on_process_finished)
self._analyzer.stage_info.connect(self._on_update_stage_info,
Qt.QueuedConnection)
self._analyzer.analyzer_error.connect(self._on_error,
Qt.QueuedConnection)
self._analyzer.start(QThread.LowestPriority)
if self._current_process_item == 0:
status('Process started...')
else:
self._abort_processing()
def _on_process_start(self, name, start_again=False):
if not self._is_running or start_again:
is_valid = True
self._is_abort = False
self._has_error = False
if self._process_items is None:
cls = self._control_buttons[name]['cls']
if type(cls) == types.ListType:
self._process_items = cls
self._current_process_item = 0
cls = cls[self._current_process_item]
# remove HmmThread if process is not first in list and
# not valid error correction was activated
if (HmmThread in self._process_items and
self._process_items.index(HmmThread) > 0 and
not (self._settings.get('Processing', 'primary_errorcorrection') or
(self._settings.get('Processing', 'secondary_errorcorrection') and
self._settings.get('Processing', 'secondary_processchannel')))):
self._process_items.remove(HmmThread)
else:
self._process_items = None
self._current_process_item = 0
else:
cls = self._process_items[self._current_process_item]
if self.SECTION_NAME == 'Classification':
result_frame = self._get_result_frame(self._tab_name)
result_frame.load_classifier(check=False)
learner = result_frame._learner
if name == self.PROCESS_PICKING:
if not result_frame.is_pick_samples():
is_valid = False
result_frame.msg_pick_samples(self)
elif result_frame.is_train_classifier():
if not question(self, 'Samples already picked',
'Do you want to pick samples again and '
'overwrite previous '
'results?'):
is_valid = False
elif name == self.PROCESS_TRAINING:
if not result_frame.is_train_classifier():
is_valid = False
result_frame.msg_train_classifier(self)
elif result_frame.is_apply_classifier():
if not question(self, 'Classifier already trained',
'Do you want to train the classifier '
'again?'):
is_valid = False
elif name == self.PROCESS_TESTING and not result_frame.is_apply_classifier():
is_valid = False
result_frame.msg_apply_classifier(self)
elif cls is HmmThread:
success, cmd = HmmThread.test_executable(self._settings.get('ErrorCorrection', 'filename_to_R'))
if not success:
critical(self, 'Error running R',
"The R command line program '%s' could not be executed.\n\n"\
"Make sure that the R-project is installed.\n\n"\
"See README.txt for details." % cmd)
is_valid = False
elif cls is MultiAnalyzerThread:
ncpu = cpu_count()
(ncpu, ok) = QInputDialog.getInt(None, "On your machine are %d processers available." % ncpu, \
"Select the number of processors", \
ncpu, 1, ncpu*2)
if not ok:
self._process_items = None
is_valid = False
if is_valid:
self._current_process = name
if not start_again:
self.parent().main_window.log_window.clear()
self._is_running = True
self._stage_infos = {}
self._toggle_tabs(False)
# disable all section button of the main widget
self.toggle_tabs.emit(self.get_name())
self._set_control_button_text(idx=1)
self._toggle_control_buttons()
imagecontainer = self.parent().main_window._imagecontainer
if cls is PickerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._set_display_renderer_info()
self._clear_image()
elif cls is AnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer)
self._set_display_renderer_info()
self._clear_image()
elif cls is TrainingThread:
self._current_settings = self._settings.copy()
self._analyzer = cls(self, self._current_settings, result_frame._learner)
self._analyzer.setTerminationEnabled(True)
self._analyzer.conf_result.connect(result_frame.on_conf_result,
Qt.QueuedConnection)
result_frame.reset()
elif cls is MultiAnalyzerThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, imagecontainer, ncpu)
self._set_display_renderer_info()
elif cls is HmmThread:
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
# FIXME: classifier handling needs revision!!!
learner_dict = {}
for kind in ['primary', 'secondary']:
_resolve = lambda x,y: self._settings.get(x, '%s_%s' % (kind, y))
env_path = CecogEnvironment.convert_package_path(_resolve('Classification', 'classification_envpath'))
if (os.path.exists(env_path)
and (kind == 'primary' or self._settings.get('Processing', 'secondary_processchannel'))
):
learner = CommonClassPredictor( \
env_path,
_resolve('ObjectDetection', 'channelid'),
_resolve('Classification', 'classification_regionname'))
learner.importFromArff()
learner_dict[kind] = learner
### Whee, I like it... "self.parent().main_window._imagecontainer" crazy, crazy, michael... :-)
self._analyzer = cls(self, self._current_settings,
learner_dict,
self.parent().main_window._imagecontainer)
self._analyzer.setTerminationEnabled(True)
lw = self.parent().main_window.log_window
lw.show()
elif cls is PostProcessingThread:
learner_dict = {}
for kind in ['primary', 'secondary']:
_resolve = lambda x,y: self._settings.get(x, '%s_%s' % (kind, y))
env_path = CecogEnvironment.convert_package_path(_resolve('Classification', 'classification_envpath'))
if (_resolve('Processing', 'classification') and
(kind == 'primary' or self._settings.get('Processing', 'secondary_processchannel'))):
learner = CommonClassPredictor( \
env_path,
_resolve('ObjectDetection', 'channelid'),
_resolve('Classification', 'classification_regionname'))
learner.importFromArff()
learner_dict[kind] = learner
self._current_settings = self._get_modified_settings(name, imagecontainer.has_timelapse)
self._analyzer = cls(self, self._current_settings, learner_dict, imagecontainer)
self._analyzer.setTerminationEnabled(True)
self._analyzer.finished.connect(self._on_process_finished)
self._analyzer.stage_info.connect(self._on_update_stage_info, Qt.QueuedConnection)
self._analyzer.analyzer_error.connect(self._on_error, Qt.QueuedConnection)
self._analyzer.start(QThread.LowestPriority)
if self._current_process_item == 0:
status('Process started...')
else:
self._abort_processing()