def main():
r = np.genfromtxt('datasets/RawData_third.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='int')
t = np.genfromtxt('datasets/RawData_time_third.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='float')
obs1 = np.zeros((r['Behaviours__1'].size, 37), dtype='int')
obs_time1 = np.zeros((t['Time__1'].size, 37), dtype='float')
n = obs1[:, 0].size
obs = np.zeros((n, 37), dtype='int')
obs.fill(-1)
obs_time = np.zeros((n, 37), dtype='float')
animalID = np.zeros(n, int)
targetID = np.zeros(n, int)
for ro in range(obs1[:, 0].size):
for col in range(36):
obs[ro][col] = r[ro][col + 5]
for row in range(obs[:, 0].size):
for col in range(36):
if (obs[row][col] == -1):
obs[row][col] = 9
for row in range(obs_time1[:, 0].size):
for col in range(36):
obs_time[row][col] = t[row][col + 5]
for row in range(obs[:, 0].size):
animalID[row] = r[row][0]
targetID[row] = r[row][4]
pos = 0
count = 0
e = 0
group1 = np.zeros(27, float)
group2 = np.zeros(27, float)
g1 = 0
g2 = 0
plot_val = np.arange(27)
error_matrix = np.zeros(54, dtype='float')
with open('Results/MPS_per_AnimalID_250_90.csv', 'w') as csvfile:
fieldnames = ['AnimalID', 'TargetID', 'PATH']
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
while pos != obs[:, 0].size:
g = 0
obs_set = np.zeros((12, 37), dtype='int')
obs_time_set = np.zeros((12, 37), dtype='float')
obs_set.fill(-1)
for i in range(pos, pos + 12):
for j in range(36):
obs_set[i - pos][j] = obs[i][j]
obs_time_set[i - pos][j] = obs_time[i][j]
T = obs_set[0].shape[0]
num_states = 9
trans_mat = hmm_train.trans_prob_matrix(obs_set)
trans_mat = np.log(trans_mat)
emi_mat_norm = hmm_train.emission_prob_matrix(obs_set)
emi_mat_norm[:, 36] = 0
emi_mat = np.log(emi_mat_norm)
emi_mat_time = hmm_train.emission_prob_matrix_time(
obs_set, obs_time_set)
emi_mat_time[:, 36] = 0
emi_mat_time = np.log(emi_mat_time)
path_set = np.empty(T, dtype='int')
path_set.fill(-1)
for t in range(T):
if (emi_mat_norm[8, t] > 0.8):
path_set[t] = -2
elif (emi_mat_norm[8, t] > 0.7 and emi_mat_norm[8, t] < 0.8):
for s in range(num_states - 1):
path_set[t] = np.argmax(emi_mat[:, t] +
trans_mat[:, s])
elif (emi_mat_norm[8, t] > 0.5 and emi_mat_norm[8, t] < 0.7):
for s in range(num_states - 1):
path_set[t] = np.argmin(emi_mat[:, t - 1] +
trans_mat[s, s])
else:
for s in range(num_states - 1):
path_set[t] = np.argmax(emi_mat[:, t - 1] +
trans_mat[s, s] +
emi_mat_time[:, t])
path_set[36] = -2
writer.writerow({
'AnimalID': str(animalID[pos]),
'TargetID': str(targetID[pos]),
'PATH': str(path_set + 1)
})
'''for r in range(37):
if(path_set[r]+1==-1):
g = g+1
if(targetID[pos]==1):
val = (36-g)/36
group1[g1] = val
g1 = g1 + 1
if(targetID[pos]==2):
val = (36-g)/36
group2[g2] = val
g2 = g2 + 1'''
pos = pos + 12
def main():
# Reading csv data file and retreiving content in variables.
r = np.genfromtxt('datasets/RawData_third.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='int')
t = np.genfromtxt('datasets/RawData_time_third.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='float')
#Declaring variables for size and storing particular data needed in analysis
obs1 = np.zeros((r['Behaviours__1'].size, 37), dtype='int')
obs_time1 = np.zeros((t['Time__1'].size, 37), dtype='float')
# Setting total size of observation sequences.
# Declaring observation sequence variables.
n = obs1[:, 0].size
obs = np.zeros((n, 37), dtype='int')
obs.fill(-1)
obs_time = np.zeros((n, 37), dtype='float')
T = obs[0].shape[0]
num_states = 9
#Populating observation sequence variables using data from four datasets.
for ro in range(obs1[:, 0].size):
for col in range(36):
obs[ro][col] = r[ro][col + 5]
for row in range(obs[:, 0].size):
for col in range(36):
if (obs[row][col] == -1):
obs[row][col] = 9
for row in range(obs_time1[:, 0].size):
for col in range(36):
obs_time[row][col] = t[row][col + 5]
# Training model based on observation sequence
#Calculating transition probability matrix based on observation
trans_mat_norm = hmm_train.trans_prob_matrix(obs)
trans_mat = np.log(trans_mat_norm)
#Calculating Emission probability matrix for model
emi_mat_norm = hmm_train.emission_prob_matrix(obs)
emi_mat_norm[:, 36] = 0
emi_mat = np.log(emi_mat_norm)
# Calculating Average time for model
emi_mat_time_norm = hmm_train.emission_prob_matrix_time(obs, obs_time)
emi_mat_time = np.log(emi_mat_time_norm)
emi_mat_time[:, 36] = 0
#All the values for transition, emission and average have log applied
#to them to avoid float underflow as probability values sometimes go very low
# Declaring variable to store most probable state sequence
path = np.empty(T, dtype='int')
path.fill(-1)
# Looping over complete observation sequence for all the states to get most
#probable state sequence
for t in range(T):
if (emi_mat_norm[8, t] > 0.8):
path[t] = -2
elif (emi_mat_norm[8, t] > 0.7 and emi_mat_norm[8, t] < 0.8):
for s in range(num_states - 1):
path[t] = np.argmax(emi_mat[:, t] + trans_mat[:, s])
elif (emi_mat_norm[8, t] > 0.5 and emi_mat_norm[8, t] < 0.7):
for s in range(num_states - 1):
path[t] = np.argmin(emi_mat[:, t - 1] + trans_mat[s, s])
else:
for s in range(num_states - 1):
path[t] = np.argmax(emi_mat[:, t - 1] + trans_mat[s, s] +
emi_mat_time[:, t])
path[36] = -2
print('==================== MOST PROBABLE STATE SEQUENCE ================')
print(path + 1)
print('==================================================================')
return (path + 1)
def main():
r = np.genfromtxt('datasets/RawData_first.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r2 = np.genfromtxt('datasets/RawData_second.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r3 = np.genfromtxt('datasets/RawData_third.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r4 = np.genfromtxt('datasets/RawData_fourth.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
t = np.genfromtxt('datasets/RawData_time_first.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t2 = np.genfromtxt('datasets/RawData_time_second.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t3 = np.genfromtxt('datasets/RawData_time_third.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t4 = np.genfromtxt('datasets/RawData_time_fourth.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
obs1 = np.zeros((r['Behaviours__1'].size,37),dtype='int')
obs1.fill(-1)
obs2 = np.zeros((r2['Behaviours__1'].size,37),dtype='int')
obs2.fill(-1)
obs3 = np.zeros((r3['Behaviours__1'].size,37),dtype='int')
obs3.fill(-1)
obs4 = np.zeros((r4['Behaviours__1'].size,37),dtype='int')
obs4.fill(-1)
obs_time1 = np.zeros((t['Time__1'].size,37),dtype='float')
obs_time2 = np.zeros((t2['Time__1'].size,37),dtype='float')
obs_time3 = np.zeros((t3['Time__1'].size,37),dtype='float')
obs_time4 = np.zeros((t4['Time__1'].size,37),dtype='float')
animalID = np.zeros(obs1[:,0].size, int)
targetID = np.zeros(obs1[:,0].size, int)
for ro in range(obs1[:,0].size):
for col in range(36):
obs1[ro][col] = r[ro][col+5]
for row in range(obs2[:,0].size):
for col in range(36):
obs2[row][col] = r2[row][col+5]
for row in range(obs3[:,0].size):
for col in range(36):
obs3[row][col] = r3[row][col+5]
for row in range(obs4[:,0].size):
for col in range(36):
obs4[row][col] = r4[row][col+5]
for row in range(obs1[:,0].size):
animalID[row] = r[row][0]
targetID[row] = r[row][4]
for row in range(obs1[:,0].size):
for col in range(36):
if(obs1[row][col]== -1):
obs1[row][col] = 9
if(obs2[row][col]== -1):
obs1[row][col] = 9
if(obs3[row][col]== -1):
obs1[row][col] = 9
if(obs4[row][col]== -1):
obs1[row][col] = 9
for row in range(obs_time1[:,0].size):
for col in range(36):
obs_time1[row][col] = t[row][col+5]
for row in range(obs_time2[:,0].size):
for col in range(36):
obs_time2[row][col] = t2[row][col+5]
for row in range(obs_time3[:,0].size):
for col in range(36):
obs_time3[row][col] = t3[row][col+5]
for row in range(obs_time4[:,0].size):
for col in range(36):
obs_time4[row][col] = t4[row][col+5]
pos = 0
error_matrix = np.zeros(54, dtype = 'float')
e=0
group1 = np.zeros(27, float)
group2 = np.zeros(27, float)
g1=0
g2=0
plot_val = np.arange(27)
with open('Results/MPS_per_AnimalID_all_Datasets.csv','w') as csvfile:
fieldnames = ['AnimalID','TargetID','PATH']
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
while pos!= obs1[:,0].size:
g = 0
obs_set = np.zeros((48,37), dtype = 'int')
obs_time_set = np.zeros((48,37), dtype='float')
obs_set.fill(-1)
for i in range(pos, pos+12):
for j in range(36):
obs_set[i-pos][j] = obs1[i][j]
obs_set[(i-pos)+12][j] = obs2[i][j]
obs_set[(i-pos)+24][j] = obs3[i][j]
obs_set[(i-pos)+36][j] = obs4[i][j]
obs_time_set[i-pos][j] = obs_time1[i][j]
obs_time_set[(i-pos)+12][j] = obs_time2[i][j]
obs_time_set[(i-pos)+24][j] = obs_time3[i][j]
obs_time_set[(i-pos)+36][j] = obs_time4[i][j]
T = obs_set[0].shape[0]
num_states = 9
trans_mat = hmm_train.trans_prob_matrix(obs_set)
trans_mat = np.log(trans_mat)
emi_mat_norm = hmm_train.emission_prob_matrix(obs_set)
emi_mat_norm[:,36] = 0
emi_mat = np.log(emi_mat_norm)
emi_mat_time = hmm_train.emission_prob_matrix_time(obs_set, obs_time_set)
emi_mat_time[:,36] = 0
emi_mat_time = np.log(emi_mat_time)
path_set = np.empty(T, dtype='int')
path_set.fill(-1)
for t in range(T):
if(emi_mat_norm[8, t] > 0.8):
path_set[t] = -2
elif(emi_mat_norm[8, t] > 0.7 and emi_mat_norm[8, t] < 0.8):
for s in range(num_states-1):
path_set[t] = np.argmax(emi_mat[:,t] + trans_mat[:, s])
elif(emi_mat_norm[8, t] > 0.5 and emi_mat_norm[8, t] < 0.7):
for s in range(num_states-1):
path_set[t] = np.argmin(emi_mat[:,t-1] + trans_mat[s, s])
else:
for s in range(num_states-1):
path_set[t] = np.argmax(emi_mat[:,t-1] + trans_mat[s,s] + emi_mat_time[:,t])
path_set[36] = -2
writer.writerow({'AnimalID' : str(animalID[pos]), 'TargetID' : str(targetID[pos]), 'PATH' : str(path_set+1)})
for r in range(37):
if(path_set[r]+1==-1):
g = g+1
if(targetID[pos]==1):
val = (36-g)/36
group1[g1] = val
g1 = g1 + 1
if(targetID[pos]==2):
val = (36-g)/36
group2[g2] = val
g2 = g2 + 1
pos = pos+12
plt.plot(plot_val, group1,'r',label="group1")
plt.plot(plot_val, group2, 'b', label="group2")
plt.legend(('group1', 'group2'))
plt.show()
def main():
r = np.genfromtxt('datasets/RawData_fourth.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='int')
t = np.genfromtxt('datasets/RawData_time_fourth.csv',
delimiter=',',
names=True,
case_sensitive=True,
dtype='float')
obs1 = np.zeros((r['Behaviours__1'].size, 37), dtype='int')
obs_time1 = np.zeros((t['Time__1'].size, 37), dtype='float')
n = obs1[:, 0].size
m = int(n / 2)
obs = np.zeros((m, 37), dtype='int')
obsg2 = np.zeros((m, 37), int)
obs.fill(-1)
obsg2.fill(-1)
obs_time = np.zeros((m, 37), dtype='float')
obsg2_time = np.zeros((m, 37), float)
T = obs[0].shape[0]
num_states = 9
for ro in range(m):
for col in range(36):
obs[ro][col] = r[ro][col + 5]
for ro in range(m, n):
for col in range(36):
obsg2[ro - m][col] = r[ro][col + 5]
for row in range(obs[:, 0].size):
for col in range(36):
if (obs[row][col] == -1):
obs[row][col] = 9
if (obsg2[row][col] == -1):
obsg2[row][col] = 9
for row in range(m):
for col in range(36):
obs_time[row][col] = t[row][col + 5]
for row in range(m, n):
for col in range(36):
obsg2_time[row - m][col] = t[row][col + 5]
trans_mat_norm1 = hmm_train.trans_prob_matrix(obs)
trans_mat1 = np.log(trans_mat_norm1)
emi_mat_norm1 = hmm_train.emission_prob_matrix(obs)
emi_mat_norm1[:, 36] = 0
emi_mat1 = np.log(emi_mat_norm1)
emi_mat_time_norm1 = hmm_train.emission_prob_matrix_time(obs, obs_time)
emi_mat_time1 = np.log(emi_mat_time_norm1)
emi_mat_time1[:, 36] = 0
trans_mat_norm2 = hmm_train.trans_prob_matrix(obsg2)
trans_mat2 = np.log(trans_mat_norm2)
emi_mat_norm2 = hmm_train.emission_prob_matrix(obsg2)
emi_mat_norm2[:, 36] = 0
emi_mat2 = np.log(emi_mat_norm2)
emi_mat_time_norm2 = hmm_train.emission_prob_matrix_time(obsg2, obsg2_time)
emi_mat_time2 = np.log(emi_mat_time_norm2)
emi_mat_time2[:, 36] = 0
path1 = np.empty(T, dtype='int')
path1.fill(-1)
path2 = np.empty(T, dtype='int')
path2.fill(-1)
for t in range(T):
if (emi_mat_norm1[8, t] > 0.8):
path1[t] = -2
elif (emi_mat_norm1[8, t] > 0.7 and emi_mat_norm1[8, t] < 0.8):
for s in range(num_states - 1):
path1[t] = np.argmax(emi_mat1[:, t] + trans_mat1[:, s])
elif (emi_mat_norm1[8, t] > 0.5 and emi_mat_norm1[8, t] < 0.7):
for s in range(num_states - 1):
path1[t] = np.argmin(emi_mat1[:, t - 1] + trans_mat1[s, s])
else:
for s in range(num_states - 1):
path1[t] = np.argmax(emi_mat1[:, t - 1] + trans_mat1[s, s] +
emi_mat_time1[:, t])
for t in range(T):
if (emi_mat_norm2[8, t] > 0.8):
path2[t] = -2
elif (emi_mat_norm2[8, t] > 0.7 and emi_mat_norm2[8, t] < 0.8):
for s in range(num_states - 1):
path2[t] = np.argmax(emi_mat2[:, t] + trans_mat2[:, s])
elif (emi_mat_norm2[8, t] > 0.5 and emi_mat_norm2[8, t] < 0.7):
for s in range(num_states - 1):
path2[t] = np.argmin(emi_mat2[:, t - 1] + trans_mat2[s, s])
else:
for s in range(num_states - 1):
path2[t] = np.argmax(emi_mat2[:, t - 1] + trans_mat2[s, s] +
emi_mat_time2[:, t])
for i in range(37):
if (path1[i] == 8):
path1[i] = -2
if (path2[i] == 8):
path2[i] = -2
path1[36] = -2
path2[36] = -2
count1 = 0
count2 = 0
for i in range(37):
if (path1[i] + 1 == -1):
count1 = count1 + 1
for i in range(37):
if (path2[i] + 1 == -1):
count2 = count2 + 1
count1 = float((36 - count1) / 36)
count2 = float((36 - count2) / 36)
print('=============== MOST PROBABLE STATE SEQUENCE =================')
print('MPS Group1 : ', path1 + 1)
print('==============================================================')
print('MPS Group2 : ', path2 + 1)
print('==============================================================')
return count1, count2, path1 + 1, path2 + 1
def main():
#Getting data from csv file and extracting in variables
r = np.genfromtxt('datasets/RawData_first.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r2 = np.genfromtxt('datasets/RawData_second.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r3 = np.genfromtxt('datasets/RawData_third.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
r4 = np.genfromtxt('datasets/RawData_fourth.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'int')
t = np.genfromtxt('datasets/RawData_time_first.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t2 = np.genfromtxt('datasets/RawData_time_second.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t3 = np.genfromtxt('datasets/RawData_time_third.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
t4 = np.genfromtxt('datasets/RawData_time_fourth.csv',delimiter=',', names=True,case_sensitive=True,dtype = 'float')
#defining observation sequences
obs1 = np.zeros((r['Behaviours__1'].size,37),dtype='int')
obs2 = np.zeros((r2['Behaviours__1'].size,37),dtype='int')
obs3 = np.zeros((r3['Behaviours__1'].size,37),dtype='int')
obs4 = np.zeros((r4['Behaviours__1'].size,37),dtype='int')
obs_time1 = np.zeros((t['Time__1'].size,37),dtype='float')
obs_time2 = np.zeros((t2['Time__1'].size,37),dtype='float')
obs_time3 = np.zeros((t3['Time__1'].size,37),dtype='float')
obs_time4 = np.zeros((t4['Time__1'].size,37),dtype='float')
#defining arrays for AnimalID and TargetID and TrialNo.
animalID = np.zeros(obs1[:,0].size, int)
targetID = np.zeros(obs1[:,0].size, int)
trialNo = np.zeros(obs1[:,0].size, int)
#Populating Observation sequence
for ro in range(obs1[:,0].size):
for col in range(36):
obs1[ro][col] = r[ro][col+5]
for row in range(obs2[:,0].size):
for col in range(36):
obs2[row][col] = r2[row][col+5]
for row in range(obs3[:,0].size):
for col in range(36):
obs3[row][col] = r3[row][col+5]
for row in range(obs4[:,0].size):
for col in range(36):
obs4[row][col] = r4[row][col+5]
for row in range(obs1[:,0].size):
animalID[row] = r[row][0]
targetID[row] = r[row][4]
trialNo[row] = r[row][2]
for row in range(obs1[:,0].size):
for col in range(36):
if(obs1[row][col]== -1):
obs1[row][col] = 9
if(obs2[row][col]== -1):
obs1[row][col] = 9
if(obs3[row][col]== -1):
obs1[row][col] = 9
if(obs4[row][col]== -1):
obs1[row][col] = 9
#Populating observtion time sequence
for row in range(obs_time1[:,0].size):
for col in range(36):
obs_time1[row][col] = t[row][col+5]
for row in range(obs_time2[:,0].size):
for col in range(36):
obs_time2[row][col] = t2[row][col+5]
for row in range(obs_time3[:,0].size):
for col in range(36):
obs_time3[row][col] = t3[row][col+5]
for row in range(obs_time4[:,0].size):
for col in range(36):
obs_time4[row][col] = t4[row][col+5]
pos = 0
error_matrix = np.zeros(54, dtype = 'float')
e=0
group1 = np.zeros((8,12), int)
group2 = np.zeros((8,12), int)
g1=0
g2=0
plot_val = np.arange(27)
#loops to write in csv file and simulatneously train and get most probable sequence
with open('Results/MPS_each_trial_all_Datasets.csv','w') as csvfile:
fieldnames = ['AnimalID','TrialNo.','TargetID','PATH']
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
while pos!= obs1[:,0].size:
g = 0
obs_set = np.zeros((4,37), dtype = 'int')
obs_time_set = np.zeros((4,37), dtype='float')
obs_set.fill(-1)
for j in range(36):
obs_set[0][j] = obs1[pos][j]
obs_set[1][j] = obs2[pos][j]
obs_set[2][j] = obs3[pos][j]
obs_set[3][j] = obs4[pos][j]
obs_time_set[0][j] = obs_time1[pos][j]
obs_time_set[1][j] = obs_time2[pos][j]
obs_time_set[2][j] = obs_time3[pos][j]
obs_time_set[3][j] = obs_time4[pos][j]
T = obs_set[0].shape[0]
num_states = 9
trans_mat = hmm_train.trans_prob_matrix(obs_set)
trans_mat = np.log(trans_mat)
emi_mat_norm = hmm_train.emission_prob_matrix(obs_set)
emi_mat_norm[:,36] = 0
emi_mat = np.log(emi_mat_norm)
emi_mat_time = hmm_train.emission_prob_matrix_time(obs_set, obs_time_set)
emi_mat_time[:,36] = 0
emi_mat_time = np.log(emi_mat_time)
path_set = np.empty(T, dtype='int')
path_set.fill(-1)
for t in range(T):
if(emi_mat_norm[8, t] > 0.8):
path_set[t] = -2
elif(emi_mat_norm[8, t] > 0.7 and emi_mat_norm[8, t] < 0.8):
for s in range(num_states-1):
path_set[t] = np.argmax(emi_mat[:,t] + trans_mat[:, s])
elif(emi_mat_norm[8, t] > 0.5 and emi_mat_norm[8, t] < 0.7):
for s in range(num_states-1):
path_set[t] = np.argmin(emi_mat[:,t-1] + trans_mat[s, s])
else:
for s in range(num_states-1):
path_set[t] = np.argmax(emi_mat[:,t-1] + trans_mat[s,s] + emi_mat_time[:,t])
path_set[36] = -2
writer.writerow({'AnimalID' : str(animalID[pos]),'TrialNo.' : str(trialNo[pos]), 'TargetID' : str(targetID[pos]), 'PATH' : str(path_set+1)})
if(targetID[pos]==1):
for i in range(37):
if(path_set[i] >= 0):
group1[path_set[i]][trialNo[pos]-1] = group1[path_set[i]][trialNo[pos]-1] + 1
if(targetID[pos]==2):
for i in range(37):
if(path_set[i] >= 0):
group2[path_set[i]][trialNo[pos]-1] = group2[path_set[i]][trialNo[pos]-1] + 1
pos = pos+1
return group1, group2