def viterbi_path(prior, transmat, obslik):
###obslik = obslik.T
scaled = 1
T = obslik.shape[1]
Q = prior.size
delta = np.zeros((Q, T))
psi = np.zeros((Q, T), dtype=int)
path = np.zeros((1, T), dtype=int)
## = np.zeros((1, T))
scale = np.ones((1, T))
t = 0
delta[:, t] = prior * obslik[:, t]
if scaled:
delta[:, t], n = normalise(delta[:, t])
scale[0, t] = 1 / n
psi[:, t] = 0
for t in range(1, T):
for j in range(0, Q):
temp = delta[:, t - 1] * transmat[:, j]
psi[j, t] = temp.argmax(axis=0)
delta[j, t] = temp[psi[j, t]] * obslik[j, t]
if scaled:
delta[:, t], n = normalise(delta[:, t])
scale[0, t] = 1 / n
###p[0, T - 1] = np.max(delta[:, T - 1])
# path[0, T - 1] = np.argmax(delta[:, T - 1]) + 1
path[0, T - 1] = np.argmax(delta[:, T - 1])
for t in range(T - 2, -1, -1):
path[0, t] = psi[(path[0, t + 1]), t + 1]
return path
def read_csv_file_by_pandas(csv_file):
data_frame = None
if (os.path.exists(csv_file)):
data_frame = pandas.read_csv(csv_file, sep='\t')
df2 = data_frame.iloc[:, 0].str.split("\t", expand=True)
for i in range(0, len(df2.index), 1):
if (df2.iloc[i, 1] != None and df2.iloc[i, 1] != ""):
indeksmaks = 0
#Normalize Severity
if (df2.iloc[i, 6] == "None" or df2.iloc[i, 6] == None
or df2.iloc[i, 6].isdigit() == False):
df2.iloc[i, 6] = 0
if (df2.iloc[i, 9] == "None" or df2.iloc[i, 9] == None
or df2.iloc[i, 9].isdigit() == False):
df2.iloc[i, 9] = 0
if (df2.iloc[i, 12] == "None" or df2.iloc[i, 12] == None
or df2.iloc[i, 12].isdigit() == False):
df2.iloc[i, 12] = 0
#Classification
if (int(df2.iloc[i, 6]) != 0 or int(df2.iloc[i, 9]) != 0
or int(df2.iloc[i, 12]) != 0):
indeksmaks = maxseverity(int(df2.iloc[i, 6]),
int(df2.iloc[i, 9]),
int(df2.iloc[i, 12]))
df2.iloc[i, 6] = int(df2.iloc[i, indeksmaks])
df2.iloc[i, 5] = "Bullying"
df2.iloc[i, 7] = df2.iloc[i, indeksmaks + 1]
else:
df2.iloc[i, 6] = 0
df2.iloc[i, 5] = "NotBullying"
df2.iloc[i, 7] = ""
df2.iloc[i, 1] = ' '.join(
normalize(
normalise(denoise_text(df2.iloc[i, 1]),
verbose=False)))
#df2.iloc[i, 7] = ' '.join(normalize(normalise(denoise_text(df2.iloc[i, 7]), verbose=False)))
else:
df2.iloc[i, 1] = "Empty"
df2.iloc[i, 6] = 0
df2.iloc[i, 5] = "NotBullying"
df2.iloc[i, 7] = ""
df2.columns = [
n.replace('', '') for n in data_frame.columns.str.split('\t')[0]
]
df2.drop("ques", axis=1, inplace=True)
df2.drop("ans", axis=1, inplace=True)
df2.drop("userid", axis=1, inplace=True)
df2.drop("asker", axis=1, inplace=True)
df2.drop("ans2", axis=1, inplace=True)
df2.drop("severity2", axis=1, inplace=True)
df2.drop("bully2", axis=1, inplace=True)
df2.drop("ans3", axis=1, inplace=True)
df2.drop("severity3", axis=1, inplace=True)
df2.drop("bully3", axis=1, inplace=True)
else:
print(csv_file + " do not exist.")
return df2
def posterior_finiteMixture(y,K,iterationsations):
# N=length(y);
a=0.1; b=0.1; alpha=5;
lambda_=np.zeros((iterations+1,K));
pi=np.zeros((iterations+1,K));
lambda_[0]=np.random.gamma(1,1,K);
pi[0]=np.random.dirichlet(alpha*np.ones((1,K))[0],1)
for i in range(1,iterations):
#z variable
p_z=-y*lambda_[i-1]+np.log(pi[i-1]*lambda_[i-1]);
p_z=normalise(p_z);
z=categorical_sample(p_z);
#lambda_ variable
n_k=sum(z);
gam_a=a+n_k; gam_b=b+sum(z*y);
lambda_[i]=np.random.gamma(gam_a,1./gam_b);
#pi variable
dir_par=alpha+n_k;
pi[i]=np.random.dirichlet(dir_par,1);
return lambda_,pi
import normalise text = ["Wat u doin ?"] normalise(text, Verbose=True)
def q_z2(y, pi, lambda_):
p_z = -y*lambda_+np.log(pi*lambda_)
p_z = normalise(p_z)
z = categorical_sample(p_z)
return z
