def pkg_dtw(file_name, line_num, df):
file_name = 'data/' + file_name
y_list = load_data(file_name, line_num)
query, reference = cal_warped_signals(y_list)
# plot warped signal
# downsample times
xvals, yinterp = get_warped_signals(query, cf.ds_time)
# calculate the corresponding point pair
query.drop(['shift', 't'], axis=1)
query2 = pd.DataFrame({'t': xvals, 'q': yinterp})
query2['close_index'] = 0
true_align_dict = get_true_aligned(cf.ds_time, query, query2)
group_num_dict = get_group_number(true_align_dict, query)
d, cost_matrix, acc_cost_matrix, path = dtw(reference[['t', 'q']].values,
query2[['t', 'q']].values,
dist=norm)
fact_align_dict = get_fact_align(path)
reverse_dict = get_reverse_dict(path)
error_rate = get_k_accuracy(true_align_dict, fact_align_dict,
group_num_dict)
SS1 = get_SS1(fact_align_dict, cf.ds_time)
SS2 = get_SS2(fact_align_dict, reverse_dict, cf.ds_time)
df.loc[line_num] = [error_rate, SS1, SS2]
return df
def event_dtw(file_name, line_num, df):
file_name = 'data/' + file_name
y_list = load_data(file_name, line_num)
query, reference = cal_warped_signals(y_list)
reference['upslope'] = 0
reference['downslope'] = 0
# plot warped signal
# downsample times
xvals, yinterp = get_warped_signals(query, cf.ds_time)
# calculate the corresponding point pair
query.drop('shift', axis=1)
query.drop('t', axis=1)
query2 = pd.DataFrame({'t': xvals, 'q': yinterp})
query2['close_index'] = 0
query2['upslope'] = 0
query2['downslope'] = 0
true_align_dict = get_true_aligned(cf.ds_time, query, query2)
group_num_dict = get_group_number(true_align_dict, query)
raw_reference_uslope, reference_upslope = get_upslope_endings(
reference['q'], cf.refer_percent)
raw_query_uslope, query_upslope = get_upslope_endings(
query2['q'], cf.query_percent)
raw_reference_downlope, reference_downslope = get_downslope_endings(
reference['q'], cf.refer_percent)
raw_query_downlope, query_downslope = get_downslope_endings(
query2['q'], cf.query_percent)
rising_edge_grps = edge_matching(reference, query2, reference_upslope,
query_upslope)
down_edge_grps = edge_matching(reference, query2, reference_downslope,
query_downslope)
calculate_event(rising_edge_grps, reference, query2, True)
calculate_event(down_edge_grps, reference, query2, False)
d, cost_matrix, acc_cost_matrix, path = dtw(
reference[['t', 'q', 'upslope', 'downslope']].values,
query2[['t', 'q', 'upslope', 'downslope']].values,
dist=norm)
fact_align_dict = get_fact_align(path)
reverse_dict = get_reverse_dict(path)
error_rate = get_k_accuracy(true_align_dict, fact_align_dict,
group_num_dict)
SS1 = get_SS1(fact_align_dict, cf.ds_time)
SS2 = get_SS2(fact_align_dict, reverse_dict, cf.ds_time)
df.loc[line_num] = [error_rate, SS1, SS2]
return df
def pkg_shapedtw(file_name, line_num, df):
file_name = 'data/' + file_name
y_list = load_data(file_name, line_num)
query, reference = cal_warped_signals(y_list)
# plot warped signal
xvals, yinterp = get_warped_signals(query, cf.ds_time)
# normalize the signal
reference_norm = stats.zscore(reference['q'])
yinterp_norm = stats.zscore(yinterp)
# store the corresponding point pair
query.drop(['shift', 't'], axis=1)
query2 = pd.DataFrame({'t': xvals, 'q': yinterp})
query2['close_index'] = 0
true_align_dict = get_true_aligned(cf.ds_time, query, query2)
group_num_dict = get_group_number(true_align_dict, query)
refer_subsequences = samplingSequences(reference_norm, sub_len)
query_subsequences = samplingSequences(yinterp_norm, int(sub_len / cf.ds_time))
refer_descriptors = np.zeros((len(refer_subsequences), nBlocks * 8))
query_descriptors = np.zeros((len(query_subsequences), nBlocks * 8))
refer_nsubsequences = len(refer_subsequences)
query_nsubsequences = len(query_subsequences)
for i in range(refer_nsubsequences):
sub_seq = refer_subsequences[i]
refer_descriptors[i] = cal_descriptor(sub_seq, sub_len)
for i in range(query_nsubsequences):
sub_seq = query_subsequences[i]
query_descriptors[i] = cal_descriptor(sub_seq, int(sub_len / cf.ds_time))
d, cost_matrix, acc_cost_matrix, path = dtw(refer_descriptors, query_descriptors, dist=norm)
fact_align_dict = get_fact_align(path)
reverse_dict = get_reverse_dict(path)
error_rate = get_k_accuracy(true_align_dict, fact_align_dict, group_num_dict)
SS1 = get_SS1(fact_align_dict, cf.ds_time)
SS2 = get_SS2(fact_align_dict, reverse_dict, cf.ds_time)
df.loc[line_num] = [error_rate, SS1, SS2]
return df
y = query['q']
yinterp = np.array(np.interp(xvals, x, y))
xvals = np.array(xvals)
ax.scatter(x=xvals, y=yinterp, marker='.', c='g', label='after interp')
ax.legend(fontsize='30')
# normalize the signal
reference_norm = stats.zscore(reference['q'])
yinterp_norm = stats.zscore(yinterp)
# store the corresponding point pair
query.drop('shift', axis=1)
query.drop('t', axis=1)
query2 = pd.DataFrame({'t': xvals, 'q2': yinterp})
query2['close_index'] = 0
true_align_dict = get_true_aligned(cf.ds_time, query, query2)
group_num_dict = get_group_number(true_align_dict, query)
query2.loc[len(query2) - 1, 'close_index'] = len(query) - 1
for i in range(len(query2) - 1):
for j in range(len(query['t2']) - 1):
if query['t2'][j] <= query2['t'][i] < query['t2'][j + 1]:
if abs(query2['q2'][i] - query['q'][j]) < abs(query2['q2'][i] -
query['q'][j + 1]):
query2.loc[i, 'close_index'] = j
else:
query2.loc[i, 'close_index'] = j + 1
if sub_len % 2 == 0:
raise Exception("Sub_len must be odd number!")
refer_subsequences = samplingSequences(reference_norm, sub_len)