def aroBW(self,iter,func,pA,pB,ppi):
inst = hmm()
alpha=func(*[self.O, pA, pB, ppi])
landa=[pA,pB,ppi,[]]
probs=np.zeros((iter,1))
P=alpha[-1]
probs[0]=[alpha[-1]]
for i in range(1,iter):
landa=inst.baum_welch(self.O, landa[0], landa[1], landa[2],5)
if (landa[-1].mean()<=P+1e-20):
landa=self.aro(100,func,landa[0], landa[1], landa[2])
P=landa[-1].mean()
probs[i]=landa[-1].max()
if (probs[i]>=1):
print(str(i)+ "inja boodeh "+ str(probs[i]))
return [landa[0], landa[1], landa[2],probs]
return [landa[0], landa[1], landa[2],probs]
def training_single(file_path, file_names):
"""
training hmm models for all 6 gestures respectively, each using one observation sequence
:param file_path: path
:param file_names: list of 6 files, one for each gesture
:return: models, a dictionary containing 6 hmm models
"""
models = {}
for i, gesture in enumerate(gesture_names):
print("\nTraining model for gesture \"{}\"...".format(gesture))
label = parse_label(gesture)
model = hmm(N[i], M) # initialize model
for file_name in file_names:
if parse_label(file_name) == label:
data = np.load(file_path+file_name)
_ = model.baum_welch(data['Z'])
models[gesture] = model
break
return models
def training_multi(file_path, file_names):
"""
training hmm models for all 6 gestures respectively, each using multiple observation sequence
:param file_path: path
:param file_names: list of 6 files, one for each gesture
:return: models, a dictionary containing 6 hmm models
"""
models = {}
for i, gesture in enumerate(gesture_names):
print("Training model for gesture \"{}\"...".format(gesture))
label = parse_label(gesture)
file_names_of_gesture = [name for name in file_names if name.startswith(gesture)]
obs_seq_set = []
model = hmm(N[i], M, use_scaling=True)
for file_name in file_names_of_gesture:
# if file_name in train_name_bank:
data = np.load(file_path + file_name)
obs_seq_set.append(data['Z'])
model.baum_welch_multi(obs_seq_set)
models[gesture] = model
return models
import gc
gc.collect()
import numpy as np
import matplotlib.pyplot as plt
from HMM import HMM as hmm
#test sequence
test_sequence = [
0, 1, 2, 1, 0, 3, 4, 4, 1, 3, 5, 5, 6, 2, 7, 8, 1, 0, 4, 7, 4, 4, 9, 5, 3,
2, 7, 10, 0, 0, 9, 6, 8, 6
]
O = np.asarray(test_sequence).reshape(len(test_sequence), 1)
inst = hmm(6, 11)
A, B, pi = inst.initializer()
iter = 500
landa_BW = inst.baum_welch(O, A, B, pi, iter)
landa_ARO = inst.aro(O, A, B, pi, iter)
landa_AROBW = inst.aroBW(O, A, B, pi, iter)
# multiple line plot
plt.plot(range(iter),
landa_BW[-1],
marker='',
markerfacecolor='blue',
markersize=3,
qlg_train_pred, post_train_pred = [], []
qlg_train_actual, post_train_actual = [], []
qlg_test_pred, post_test_pred = [], []
qlg_test_actual, post_test_actual = [], []
numkc = 10 # change number of kc for different data set
for train_index, test_index in kf.split(qlg_y):
qlg_y_train, qlg_y_test = qlg_y[train_index], qlg_y[test_index]
post_y_train, post_y_test = post_y[train_index], post_y[test_index]
# symbols here refers to four differnt observations:
# fast-correct, fast-incorrect, slow-correct, slow-incorrect
symbols = [['0', '1', '2', '3']]
h = hmm(2,
Pi=np.array([0.8, 0.2]),
T=np.array([[0.8, 0.2], [0.1, 0.9]]),
obs_symbols=symbols)
# change parameters here F= T=
nlg_train = [[] for x in range(numkc)]
nlg_test = [[] for x in range(numkc)]
for i in range(numkc):
X = np.load(folder + "new_bkt_pertime_kc" + str(i) + ".npy")
X_train, X_test = X[train_index], X[test_index]
train = [each for each in X_train if each]
test = [each for each in X_test if each]
if train and test:
h.baum_welch(train, debug=False) # training part
nlg_train[i].extend(h.predict_nlg(X_train))
matrizIntermediate_train = matrizIntermediate[train_index]
matrizIntermediate_test = matrizIntermediate[test_index]
# Simbolos: 1 incorrecto, 2 correcto
symbols = [['1', '2']]
#Objeto HMM pasando como parámetro:
# Probabilidad Inicial PI --> P(L0) 1-P(L0)
# Calculada en base al número de aciertos de los puzzles de tutorial
# Probabilidad de cambiar de un estado a otro T --> A
# Probabilidad de Emission E --> B
# Símbolos anteriores
h = hmm(2,
Pi=np.array([0.48, 0.52]),
T=np.array([[1, 0], [0.4, 0.6]]),
E=[np.array([[0.95, 0.05], [0.05, 0.95]])],
obs_symbols=symbols)
#h = hmm(2, Pi=np.array([0.8, 0.2]), T=np.array([[1, 0], [0.4, 0.6]]), obs_symbols=symbols)
#Inicialización a vacío
conjunto_train = [[] for x in range(numkc)]
conjunto_test = [[] for x in range(numkc)]
#Iteración para cada kc, en este caso 1 ya que no se ha hecho diferenciación de puzzles
for i in range(numkc):
#Matriz con lista de respuestas por alumno
datosPuzzles = np.load(folder + "IntermediatePuzzlesResults.npy",
allow_pickle=True)
print(datosPuzzles)