def make_train():
#posi_tempo=[]
print "start positive music feature extract"
posi_mel=[]
posi_chroma=[]
posi_mfcc=[]
hi_res=0
#posi_tempo=[]
#nega_tempo=[]
with warnings.catch_warnings():
warnings.simplefilter("ignore")
for i in tqdm(positive_sample):
#tools.make_vec(os.path.join(posi_dir,i),0,(posi_mel,posi_mfcc,posi_chroma))
try:
mel,mfcc,chroma = tools.get_feature(os.path.join(posi_dir,i))
mel=[(0,j) for j in mel]
mfcc=[(0,j) for j in mfcc]
chroma=[(0,j) for j in chroma]
posi_mel+=mel
posi_mfcc+=mfcc
posi_chroma+=chroma
tools.make_spec(os.path.join(posi_dir,i),os.path.join(data_dir,"posi_spectro",i[:-3]+"png"))
except audioop.error:
hi_res+=1
p_dump(posi_mel,"posi_mel")
p_dump(posi_mfcc,"posi_mfcc")
p_dump(posi_chroma,"posi_chroma")
del posi_mel
del posi_mfcc
del posi_chroma
print "start negative music feature extract"
nega_mel=[]
nega_chroma=[]
nega_mfcc=[]
for i in tqdm(negative_sample):
#tools.make_vec(os.path.join(nega_dir,i),0,(nega_mel,nega_mfcc,nega_chroma))
try:
mel,mfcc,chroma=tools.get_feature(os.path.join(nega_dir,i))
mel=[(1,j) for j in mel]
mfcc=[(1,j) for j in mfcc]
chroma=[(1,j) for j in chroma]
nega_mel+=mel
nega_mfcc+=mfcc
nega_chroma+=chroma
tools.make_spec(os.path.join(nega_dir,i),os.path.join(data_dir,"nega_spectro",i[:-3]+"png"))
except audioop.error:
hi_res+=1
p_dump(nega_mel,"nega_mel")
p_dump(nega_mfcc,"nega_mfcc")
p_dump(nega_chroma,"nega_chroma")
if hi_res:
print "{} files are not used.it may be 24bit.".format(hi_res)
def get_features_values_by_sampling(graph):
iterations = 1000
features = np.zeros((iterations, 7))
for iteration in range(iterations):
g = graph.copy()
g = tools.create_random_graph(g)
features[iteration, :] = tools.get_feature(g)
return features
def get_features_values_by_competition(graph):
iterations = run.feature_iterations
features = np.zeros((iterations * 1600, 7))
counter = 0
for iteration in range(iterations):
g = graph.copy()
while len(g.graph['free']) > 1:
feature = tools.get_feature(g)
features[counter][:] = feature
counter += 1
# first player action
action_choice1 = np.random.randint(3)
if action_choice1 == 0:
seed1 = actions.action_degree(g)
elif action_choice1 == 1:
seed1 = actions.action_weight(g)
elif action_choice1 == 2:
seed1 = actions.action_blocking(g, 1)
# illegal action
if seed1 == -1:
print("Illegal action")
action_choice1 = 0
seed1 = actions.action_degree(g)
tools.activate_node(g, seed1, 1)
# second player action
action_choice2 = np.random.randint(3)
if action_choice2 == 0:
seed2 = actions.action_degree(g)
elif action_choice2 == 1:
seed2 = actions.action_weight(g)
elif action_choice2 == 2:
seed2 = actions.action_blocking(g, 2)
# illegal action
if seed2 == -1:
print("Illegal action")
action_choice2 = 0
seed2 = actions.action_degree(g)
tools.activate_node(g, seed2, 2)
a1, a2 = tools.diffuse(g)
for n in a1:
tools.activate_node(g, n, 1)
for n in a2:
tools.activate_node(g, n, 2)
return features[0:counter, :]
def param_coord_mapping_Ver2(sampling=200):
from get_indices import finished_modify
from tools import get_feature
from core import G
import humanmodifier
with open("/home/loaias/Workspace/Python/MakeHuman/log/parts/indices", 'r') as f:
indices = pickle.load(f)
mapping=[
("eye_width", ["eyes/l-eye-push1-in|out", "eyes/l-eye-push2-in|out"]),
("eye_height", "eyes/l-eye-height2-min|max"),
("eye_distance", "eyes/l-eye-move-in|out"),
("face_width", "head/head-scale-horiz-less|more"),
("face_height", "chin/chin-height-min|max"),
("nose_width", "nose/nose-scale-horiz-incr|decr"),
("forehead_to_nose", "nose/nose-scale-vert-incr|decr"),
("nose_to_mouth", "mouth/mouth-trans-up|down")
]
h = G.app.selectedHuman
params = np.linspace(-1, 1, sampling)
for el in mapping:
coord = []
for param in params:
if len(el[1]) == 2:
modifier = h.getModifier(el[1][0])
humanmodifier.ModifierAction(modifier, 0, param, finished_modify).do()
modifier = h.getModifier(el[1][1])
humanmodifier.ModifierAction(modifier, 0, -param, finished_modify).do()
else:
modifier = h.getModifier(el[1])
humanmodifier.ModifierAction(modifier, 0, param, finished_modify).do()
mesh = h.mesh.__dict__["coord"]
coord.append(get_feature(el[0], mesh))
result = [params, coord]
np.save("%s/mapping_ver2/%s.npy" % (base_path, el[0]), result)
raw_input("Parameter-Coordinate Mapping Complete.")
def main(m_list):
mel_model = models.Mel()
mfcc_model = models.Mfcc()
chroma_model = models.Chroma()
spec_model = models.Spectro()
law_model = models.Law()
serializers.load_hdf5(os.path.join(model_dir, "my_mel.model"), mel_model)
serializers.load_hdf5(os.path.join(model_dir, "my_mfcc.model"), mfcc_model)
serializers.load_hdf5(os.path.join(model_dir, "my_chroma.model"),
chroma_model)
serializers.load_hdf5(os.path.join(model_dir, "my_law.model"), law_model)
serializers.load_hdf5(os.path.join(model_dir, "my_spec.model"), spec_model)
candidate = []
with warnings.catch_warnings():
warnings.simplefilter("ignore")
for i in tqdm(m_list):
try:
p_n = []
#print "begin",i,"({0}/{1})".format(num+1,len(m_list))
mel, mfcc, chroma = tools.get_feature(i)
p_n.append(predict_mmc(mel_model, mel).mean(0))
#tools.fprint("mel done ")
p_n.append(predict_mmc(mfcc_model, mfcc).mean(0))
#tools.fprint("mfcc done ")
p_n.append(predict_mmc(chroma_model, chroma).mean(0))
#tools.fprint("chroma done")
p_n.append(predict_law(law_model, i).mean(0))
#tools.fprint("law done ")
p_n.append(predict_spec(spec_model, i))
#tools.fprint("spec done ")
candidate.append((i, p_n))
#tools.fprint("end \n")
except audioop.error:
pass
return candidate
def main(m_list):
mel_model = models.Mel()
mfcc_model = models.Mfcc()
chroma_model = models.Chroma()
spec_model = models.Spectro()
law_model = models.Law()
serializers.load_hdf5(os.path.join(model_dir,"my_mel.model"), mel_model)
serializers.load_hdf5(os.path.join(model_dir,"my_mfcc.model"), mfcc_model)
serializers.load_hdf5(os.path.join(model_dir,"my_chroma.model"), chroma_model)
serializers.load_hdf5(os.path.join(model_dir,"my_law.model"), law_model)
serializers.load_hdf5(os.path.join(model_dir,"my_spec.model"), spec_model)
candidate = []
with warnings.catch_warnings():
warnings.simplefilter("ignore")
for i in tqdm(m_list):
try:
p_n=[]
#print "begin",i,"({0}/{1})".format(num+1,len(m_list))
mel,mfcc,chroma = tools.get_feature(i)
p_n.append(predict_mmc(mel_model,mel).mean(0))
#tools.fprint("mel done ")
p_n.append(predict_mmc(mfcc_model,mfcc).mean(0))
#tools.fprint("mfcc done ")
p_n.append(predict_mmc(chroma_model,chroma).mean(0))
#tools.fprint("chroma done")
p_n.append(predict_law(law_model,i).mean(0))
#tools.fprint("law done ")
p_n.append(predict_spec(spec_model,i))
#tools.fprint("spec done ")
candidate.append((i,p_n))
#tools.fprint("end \n")
except audioop.error:
pass
return candidate
def param_coord_mapping_Ver3(sampling=20):
from get_indices import finished_modify, get_feature
from core import G
import humanmodifier
# with open("/home/loaias/Workspace/Python/MakeHuman/log/parts/indices", 'r') as f:
# indices = pickle.load(f)
mapping = [
("eye-out", "eyes/l-eye-push1-in|out"),
("eye-in", "eyes/l-eye-push2-in|out"),
("eye-height", "eyes/l-eye-height2-min|max"),
("eye-distance", "eyes/l-eye-move-in|out"),
("face-width", "head/head-scale-horiz-less|more"),
# ("face_height", "chin/chin-height-min|max"),
("face-height", "head/head-scale-vert-more|less"),
("nose-width", "nose/nose-scale-horiz-incr|decr"),
("nose-apex", "nose/nose-scale-vert-incr|decr"),
("mouth-middle", "mouth/mouth-trans-up|down")
]
h = G.app.selectedHuman
params = np.linspace(-1, 1, sampling)
for el in mapping:
parameter_feature_pairs = []
for param in params:
modifier = h.getModifier(el[1])
humanmodifier.ModifierAction(modifier, 0, param, finished_modify).do()
mesh = h.mesh.__dict__["coord"]
feature = get_feature(el[0], mesh)
parameter_feature_pairs.append((param, feature))
np.save("%s/mapping_ver3/%s.npy" % (base_path, el[0]), parameter_feature_pairs)
print("Parameter-Coordinate Mapping Complete.")
def make_train():
#posi_tempo=[]
print "start positive music feature extract"
posi_mel = []
posi_chroma = []
posi_mfcc = []
hi_res = 0
#posi_tempo=[]
#nega_tempo=[]
with warnings.catch_warnings():
warnings.simplefilter("ignore")
for i in tqdm(positive_sample):
#tools.make_vec(os.path.join(posi_dir,i),0,(posi_mel,posi_mfcc,posi_chroma))
try:
mel, mfcc, chroma = tools.get_feature(os.path.join(
posi_dir, i))
mel = [(0, j) for j in mel]
mfcc = [(0, j) for j in mfcc]
chroma = [(0, j) for j in chroma]
posi_mel += mel
posi_mfcc += mfcc
posi_chroma += chroma
tools.make_spec(
os.path.join(posi_dir, i),
os.path.join(data_dir, "posi_spectro", i[:-3] + "png"))
except audioop.error:
hi_res += 1
p_dump(posi_mel, "posi_mel")
p_dump(posi_mfcc, "posi_mfcc")
p_dump(posi_chroma, "posi_chroma")
del posi_mel
del posi_mfcc
del posi_chroma
print "start negative music feature extract"
nega_mel = []
nega_chroma = []
nega_mfcc = []
for i in tqdm(negative_sample):
#tools.make_vec(os.path.join(nega_dir,i),0,(nega_mel,nega_mfcc,nega_chroma))
try:
mel, mfcc, chroma = tools.get_feature(os.path.join(
nega_dir, i))
mel = [(1, j) for j in mel]
mfcc = [(1, j) for j in mfcc]
chroma = [(1, j) for j in chroma]
nega_mel += mel
nega_mfcc += mfcc
nega_chroma += chroma
tools.make_spec(
os.path.join(nega_dir, i),
os.path.join(data_dir, "nega_spectro", i[:-3] + "png"))
except audioop.error:
hi_res += 1
p_dump(nega_mel, "nega_mel")
p_dump(nega_mfcc, "nega_mfcc")
p_dump(nega_chroma, "nega_chroma")
if hi_res:
print "{} files are not used.it may be 24bit.".format(hi_res)
knn = KNeighborsClassifier(n_neighbors=neighbors)
knn.fit(train_data, train_label)
# predict_labe = knn.predict(test_data)
score = knn.score(test_data, test_label)
print("精度为： ", score)
# tools.tsen_plot("KNeighbor",features,labels)
return score
if __name__ == "__main__":
simi_path = "E:/Program Files/workspace/report_sheng/report_similarity/"
features, labels, words, feature_names = tools.get_feature(simi_path)
labels_list = []
for i in range(len(labels)):
labels_list.append(lable_dict[labels[i]])
word_matrix, word_names = tools.get_count_vect(words)
score_dict = {}
for num in trange(10, 1137):
#使用卡方分布从词频特征抽取特征
new_feature = tools.select_feature(word_matrix.toarray(), labels_list,
num)
for neibhor in range(3, 32):
#多项式贝叶斯分类
