def get_hmms(self):
for gesture_type in self.gesture_types:
print_status("Get_Hmms",
"Fitting for gesture_type: " + gesture_type)
### Step 1: fill hmm_examples appropriately ###
hmm_examples = []
for gesture in self.gestures[gesture_type]:
hmm_rep = gesture.get_hmm_rep()
hmm_examples.append(hmm_rep)
### Step 2: fit parameters for the hmm ###
hmm = GaussianHMM(self.num_hmm_states)
hmm.fit(hmm_examples)
### Step 3: store the hmm in self.hmms ###
self.hmms[gesture_type] = hmm
print_inner_status(
gesture_type,
"predicted the following sequences: (score: sequence)")
for example in hmm_examples:
print " ", hmm.score(example), ": ", hmm.predict(example)
def get_correspondences(self):
print_inner_status("Get correspondences", "getting the intervals")
### Step 1: get all of the beat indeces, pop indeces ###
self.beat_indeces = [
i for i in range(len(self.original_skeletons))
if self.original_skeletons[i].beat == 1
]
beat_intervals = []
### Step 2: get a list of beat intervals ###
prev_beat = 0
for i in range(len(self.beat_indeces) - 1):
cur_beat = self.beat_indeces[i]
next_beat = self.beat_indeces[i + 1]
beat_interval = Beat_Interval(cur_beat, prev_beat, next_beat)
beat_intervals.append(beat_interval)
prev_beat = cur_beat
### for each beat_interval, iterate through each skeleton and select the top prob ###
self.correspondences = []
for beat_interval in beat_intervals:
best_prob = -1.0
best_index = -1
best_class = -1
for i in range(beat_interval.interval_start,
beat_interval.interval_end):
pop_probability = self.original_skeletons[i].pop_probability
print i, ": ", pop_probability
# for index, p in enumerate(pop_probability[1:]): #use this for binary classification
for index, p in enumerate(
pop_probability[:-1]
): #use this for multiple-class classification
if p > best_prob:
best_index = i
best_prob = p
best_class = index
self.correspondences.append((beat_interval.beat_index, best_index))
print "### best_prob: ", best_prob
print "### best_index: ", best_index
print "### best_class: ", best_class
### Success up to here - correspondences is now filled with (beat, best pop) index pairs ###
### Now mark beat/pop and write it out
self.marked_skeletons = self.original_skeletons
self.pop_indeces = [i[1] for i in self.correspondences]
for i in range(len(self.marked_skeletons)):
if i in set(self.pop_indeces):
self.marked_skeletons[i].pop = 1
else:
self.marked_skeletons[i].pop = 0
def get_correspondences (self):
print_inner_status ("Get correspondences", "getting the intervals")
### Step 1: get all of the beat indeces, pop indeces ###
self.beat_indeces = [i for i in range(len(self.original_skeletons)) if self.original_skeletons[i].beat == 1]
beat_intervals = []
### Step 2: get a list of beat intervals ###
prev_beat = 0
for i in range (len(self.beat_indeces) - 1):
cur_beat = self.beat_indeces [i]
next_beat = self.beat_indeces [i + 1]
beat_interval = Beat_Interval (cur_beat, prev_beat, next_beat)
beat_intervals.append (beat_interval)
prev_beat = cur_beat
### for each beat_interval, iterate through each skeleton and select the top prob ###
self.correspondences = []
for beat_interval in beat_intervals:
best_prob = -1.0
best_index = -1
best_class = -1
for i in range (beat_interval.interval_start, beat_interval.interval_end):
pop_probability = self.original_skeletons[i].pop_probability
print i, ": ", pop_probability
# for index, p in enumerate(pop_probability[1:]): #use this for binary classification
for index, p in enumerate (pop_probability[:-1]): #use this for multiple-class classification
if p > best_prob:
best_index = i
best_prob = p
best_class = index
self.correspondences.append ((beat_interval.beat_index, best_index))
print "### best_prob: ", best_prob
print "### best_index: ", best_index
print "### best_class: ", best_class
### Success up to here - correspondences is now filled with (beat, best pop) index pairs ###
### Now mark beat/pop and write it out
self.marked_skeletons = self.original_skeletons
self.pop_indeces = [i[1] for i in self.correspondences]
for i in range(len(self.marked_skeletons)):
if i in set(self.pop_indeces):
self.marked_skeletons[i].pop = 1
else:
self.marked_skeletons[i].pop = 0
def save_examples (self):
print_status ("Save Examples", "Begin")
### Step 1: save self.all_examples ###
for meme_type, meme_list in self.all_examples.iteritems ():
save_filename = self.get_examples_pickle_filename (meme_type)
print_inner_status ("Save Examples", "Saving to " + save_filename)
pickle.dump (meme_list, open(save_filename, 'w'))
print_status ("Save Examples", "End")
def get_examples (self):
### Step 1: fill self.all_examples ###
self.all_examples = {}
print_status ("Get Examples", "Begin")
for meme_type, meme_list in self.memes.iteritems():
print_inner_status ("Get Examples", "Converting to feature vectors: " + meme_type)
self.all_examples [meme_type] = [m.get_features () for m in meme_list]
print_status ("Get Examples", "Complete")
def load_memes (self):
print_status ("Loading memes", "Begin")
meme_pkl_filenames = [f for f in os.listdir (self.memes_directory) if f[-4:] == '.pkl']
for meme_pkl_filename in meme_pkl_filenames:
full_filename = os.path.join (self.memes_directory, meme_pkl_filename)
meme_type = meme_pkl_filename[:-4]
print_inner_status ("Load Memes", "Loading " + full_filename)
self.memes[meme_type] = pickle.load (open(full_filename, 'r'))
print_status ("Loading memes", "End")
def load_examples (self):
print_status ("Load Examples", "Begin")
self.all_examples = {}
example_filenames = [f for f in os.listdir (self.examples_directory) if f[-12:] == '.feature_vec']
for example_filename in example_filenames:
full_filename = os.path.join (self.examples_directory, example_filename)
meme_type = example_filename[:-12]
print_inner_status ("Load Examples", "Loading " + meme_type)
self.all_examples [meme_type] = pickle.load(open(full_filename, 'r'))
print_status ("Load Examples", "Complete")
def load_gestures (self):
### Step 1: initialize self.gestures ###
self.gestures = {}
### Step 2: get all gesture types ###
self.get_gesture_types ()
### Step 3: get all gesture dirs ###
self.get_gesture_dirs ()
### Step 4: for each gesture_type, load in all gestures ###
for gesture_type in self.gesture_types:
print_inner_status ("Gesture Recognizer (load_gestures)", "Loading gestures of type " + str(gesture_type))
self.load_gestures_of_type (gesture_type)
def load_gestures(self):
### Step 1: initialize self.gestures ###
self.gestures = {}
### Step 2: get all gesture types ###
self.get_gesture_types()
### Step 3: get all gesture dirs ###
self.get_gesture_dirs()
### Step 4: for each gesture_type, load in all gestures ###
for gesture_type in self.gesture_types:
print_inner_status("Gesture Recognizer (load_gestures)",
"Loading gestures of type " + str(gesture_type))
self.load_gestures_of_type(gesture_type)
def get_hmms (self):
for gesture_type in self.gesture_types:
print_status ("Get_Hmms", "Fitting for gesture_type: " + gesture_type)
### Step 1: fill hmm_examples appropriately ###
hmm_examples = []
for gesture in self.gestures[gesture_type]:
hmm_rep = gesture.get_hmm_rep ()
hmm_examples.append (hmm_rep)
### Step 2: fit parameters for the hmm ###
hmm = GaussianHMM (self.num_hmm_states)
hmm.fit (hmm_examples)
### Step 3: store the hmm in self.hmms ###
self.hmms[gesture_type] = hmm
print_inner_status (gesture_type, "predicted the following sequences: (score: sequence)")
for example in hmm_examples:
print " ", hmm.score (example), ": ", hmm.predict (example)
def __init__ (self, sync_directory, classifier_name):
### Step 1: filename management ###
print_inner_status ("Initialization", "Looking at file " + sync_directory)
self.jvid_filename_raw = os.path.join(sync_directory, 'Raw/video.jvid')
self.jvid_filename_pops_marked = os.path.join(sync_directory, 'Marked/video.jvid')
self.jvid_filename_synchronized = os.path.join(sync_directory, 'Synced/video.jvid')
### Step 2: load the classifier ###
print_inner_status ("Initialization", "Unpickling the classifier")
self.classifier = pickle.load (open('/Users/jhack/Programming/NI/ni_template/python_backend/classifiers/toprock_front_training.obj', 'r'))
### Step 2: get the original skeletons ###
print_inner_status ("Initialization", "Getting input skeletons")
self.original_skeletons = read_in_skeletons (self.jvid_filename_raw)
### Step 3: add derivatives to them ###
print_inner_status ("Initialization", "Adding derivatives to skeletons")
self.original_skeletons = add_derivatives_to_skeletons(self.original_skeletons, 5, 10, 5, 10)
### Step 4: add pop probabilities to them ###
print_inner_status ("Initialization", "Adding pop probabilities to skeletons")
self.original_skeletons = mark_pop_probabilities (self.original_skeletons, self.classifier)
def __init__(self, sync_directory, classifier_name, mode):
### Step 1: filename management ###
print_inner_status("Initialization",
"Looking at file " + sync_directory)
self.jvid_filename_raw = os.path.join(sync_directory, 'Raw/video.jvid')
self.jvid_filename_pops_marked = os.path.join(sync_directory,
'Marked/video.jvid')
self.jvid_filename_synchronized = os.path.join(sync_directory,
'Synced/video.jvid')
### Step 2: get the original skeletons ###
print_inner_status("Initialization", "Getting input skeletons")
self.original_skeletons = read_in_skeletons(self.jvid_filename_raw)
### Step 3: add derivatives to them ###
print_inner_status("Initialization", "Adding derivatives to skeletons")
self.original_skeletons = add_derivatives_to_skeletons(
self.original_skeletons, 5, 10, 5, 10)
### Step 4: set up the classifier filename ###
self.classifier_filename = os.path.join(
os.getcwd(), 'python_backend/classifiers/' + classifier_name)
### Step 5.0: if synchronize mode, load the classifier and mark probabilities ###
if mode == 'synchronize':
print_inner_status("Initialization", "Loading the classifier")
self.load_classifier(
'/Users/jhack/Programming/NI/ni_template/python_backend/classifiers/toprock_front_training.obj'
)
print_inner_status("Initialization",
"Adding pop probabilities to skeletons")
self.original_skeletons = mark_pop_probabilities(
self.original_skeletons, self.classifier)
### Step 5.1: if train mode, train the classifier and save it ###
elif mode == 'train':
print_inner_status("Initialization", "Training the classifier")
self.train_classifier()
print_inner_status("Initialization", "Saving the classifier")
self.save_classifier()
return
def __init__ (self, sync_directory, classifier_name, mode):
### Step 1: filename management ###
print_inner_status ("Initialization", "Looking at file " + sync_directory)
self.jvid_filename_raw = os.path.join(sync_directory, 'Raw/video.jvid')
self.jvid_filename_pops_marked = os.path.join(sync_directory, 'Marked/video.jvid')
self.jvid_filename_synchronized = os.path.join(sync_directory, 'Synced/video.jvid')
### Step 2: get the original skeletons ###
print_inner_status ("Initialization", "Getting input skeletons")
self.original_skeletons = read_in_skeletons (self.jvid_filename_raw)
### Step 3: add derivatives to them ###
print_inner_status ("Initialization", "Adding derivatives to skeletons")
self.original_skeletons = add_derivatives_to_skeletons(self.original_skeletons, 5, 10, 5, 10)
### Step 4: set up the classifier filename ###
self.classifier_filename = os.path.join (os.getcwd(), 'python_backend/classifiers/' + classifier_name)
### Step 5.0: if synchronize mode, load the classifier and mark probabilities ###
if mode == 'synchronize':
print_inner_status ("Initialization", "Loading the classifier")
self.load_classifier ('/Users/jhack/Programming/NI/ni_template/python_backend/classifiers/toprock_front_training.obj')
print_inner_status ("Initialization", "Adding pop probabilities to skeletons")
self.original_skeletons = mark_pop_probabilities (self.original_skeletons, self.classifier)
### Step 5.1: if train mode, train the classifier and save it ###
elif mode == 'train':
print_inner_status ("Initialization", "Training the classifier")
self.train_classifier ()
print_inner_status ("Initialization", "Saving the classifier")
self.save_classifier ()
return
