def time_corrDim(root):
'''
Creates plots of path length vs area per unit length
'''
df = idf[idf['Length'] > 20][idf['Point Count'] > 20][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(1000)
df = df.reset_index(drop=True)
times = []
corrDims = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
times.append(t.time)
corrDims.append(t.corrDim())
l = np.array([corrDims, times])
plt.scatter(l[0], l[1], s=1)
plt.ylabel('Duration (minutes)')
plt.xlabel('Correlation Dimension')
plt.show()
return l
def hurst_area(root, samples=500):
'''
Creates plot of hurst exponent vs area covered per unit length by a trajectory
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
hursts = []
areas = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
hursts.append(t.hurst())
areas.append(t.areaPerUnitL())
l = np.array([areas, hursts])
plt.scatter(l[0], l[1], s=1)
plt.xlabel('Area Covered per unit length (m)')
plt.ylabel('Mean Hurst exponent')
plt.show()
return l
def area_time(root, samples=500):
'''
Creates plot of area covered vs duration of a trajectory
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
areas = []
times = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
times.append(t.time * 60)
areas.append(t.coveredArea())
l = np.array([times, areas])
slope, intercept, r_value, p_value, std_err = stats.linregress(l[0], l[1])
plt.scatter(l[0], l[1], s=1)
plt.plot(l[0], intercept + slope * l[0], 'r', label='fitted line')
plt.xlabel('Duration (s)')
plt.ylabel('Area Covered (m^2)')
# plt.show()
return l
def hurst_length(root):
'''
Creates plots of hurst exponent vs length of trajectory
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(1000)
df = df.reset_index(drop=True)
hursts = []
lengths = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
hursts.append(t.hurst())
lengths.append(t.len)
l = np.array([lengths, hursts])
plt.scatter(l[0], l[1], s=1)
plt.xlabel('Length (m)')
plt.ylabel('Mean Hurst exponent')
plt.show()
return l
def length_area(root):
'''
Creates plots of path length vs area per unit length
'''
df = idf[idf['Length'] > 20][idf['Point Count'] > 20][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(1000)
df = df.reset_index(drop=True)
lengths = []
areaPerUnitLs = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
lengths.append(t.len)
areaPerUnitLs.append(t.areaPerUnitL())
l = np.array([lengths, areaPerUnitLs])
plt.scatter(l[0], l[1], s=1)
plt.xlabel('Length (m)')
plt.ylabel('Area per unit length (m)')
plt.show()
return l
def hurst_XY(root):
'''
Creates plots of path the hurst exponent in the y-direction vs x-direction
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(1000)
df = df.reset_index(drop=True)
hurst_x = []
hurst_y = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
hurst = t.hurst(xy=True)
hurst_x.append(hurst[0])
hurst_y.append(hurst[1])
l = np.array([hurst_x, hurst_y])
plt.scatter(l[0], l[1], s=1)
plt.xlabel('Hurst exponent in x-direction')
plt.ylabel('Hurst exponent in y-direction')
plt.show()
return l
def length_corrDim(root):
'''
Creates plots of length vs correlation dimension
'''
df = idf[idf['Length'] > 20][idf['Point Count'] > 50][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(500)
df = df.reset_index(drop=True)
lengths = []
corrDims = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
lengths.append(t.len)
corrDims.append(t.corrDim())
l = np.array([lengths, corrDims])
plt.scatter(l[0], l[1], s=1)
plt.xlabel('Length (m)')
plt.ylabel('Correlation Dimension')
plt.show()
return l
def angleS_mode(root, samples=500):
'''
Creates plot of turning angle per unit length vs mode of transport
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 20][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
angles = [[], [], [], [], [], [], [], [], [], [], [], []]
modes = [
'walk', 'run', 'car', 'train', 'airplane', 'taxi', 'bus', 'subway',
'bike', 'boat', 'motorcycle', 'Unlabelled'
]
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
i = modes.index(t.mode)
angles[i].append(t.angleDensS())
l = np.array([[np.median(x), iqr(x)] for x in angles])
N = len(modes)
fig, ax = plt.subplots()
ind = np.arange(N) # the x locations for the groups
width = 0.35 # the width of the bars
fig, ax = plt.subplots()
ax.boxplot(angles)
# add some text for labels, title and axes ticks
ax.set_ylabel('Turning Angle Density (degrees/metre)')
ax.set_xticks(ind + 1)
ax.set_xticklabels(modes)
# Rotate mode labels to a slant
for label in ax.get_xmajorticklabels():
label.set_rotation(50)
label.set_horizontalalignment("right")
plt.xlabel('Mode of Transport')
# plt.show()
return l
def area_length(root, samples=500):
'''
Creates plot of window area vs length of a trajectory
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
areas = []
lengths = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
lengths.append(t.len)
areas.append(t.windowArea())
# areas.append(t.coveredArea(radius=500))
l = np.array([lengths, areas])
coeffs, cov = np.polyfit(np.log(lengths), np.log(areas), 1, cov=True)
error = np.sqrt(np.diag(cov))
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.scatter(l[0], l[1], s=1)
ax.set_yscale('log')
ax.set_xscale('log')
x = np.geomspace(np.min(l[0]), np.max(l[0]), 100)
ax.plot(x, np.exp(coeffs[1]) * x**coeffs[0], 'r', linewidth=0.7)
ax.set_xlabel('Path Length (m)')
ax.set_ylabel('Window Area ($m^2$)')
ax.set_title('Exponent: ' + str(coeffs[0]) + '$\pm$' + str(error[0]))
# plt.show()
return [coeffs[0], error[0]]
def corrDim_mode(root, samples=500):
'''
Creates plot of correlation dimension vs mode of transport
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 60][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
dims = [[], [], [], [], [], [], [], [], [], [], [], []]
modes = [
'walk', 'run', 'car', 'train', 'airplane', 'taxi', 'bus', 'subway',
'bike', 'boat', 'motorcycle', 'Unlabelled'
]
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
i = modes.index(t.mode)
dims[i].append(t.corrDim())
dims = [[dim for dim in mode if not np.isnan(dim)] for mode in dims]
l = np.array([[np.median(x), iqr(x)] for x in dims])
N = len(modes)
ind = np.arange(N) # the x locations for the groups
fig, ax = plt.subplots()
ax.boxplot(dims)
# errors = ax.errorbar(ind, l[:,0], yerr=l[:,1], fmt='x', capsize=2)
# add some text for labels, title and axes ticks
ax.set_ylabel('Correlation dimension')
ax.set_xticks(ind + 1)
ax.set_xticklabels(modes, rotation='vertical')
plt.xlabel('Mode of Transport')
plt.show()
return dims
def areaT_mode(root, samples=500):
'''
Creates plot of area covered per unit time vs mode of transport
'''
df = idf[idf['Length'] > 20][idf['Point Count'] >= 20][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(samples)
df = df.reset_index(drop=True)
areas = [[], [], [], [], [], [], [], [], [], [], [], []]
modes = [
'walk', 'run', 'car', 'train', 'airplane', 'taxi', 'bus', 'subway',
'bike', 'boat', 'motorcycle', 'Unlabelled'
]
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
i = modes.index(t.mode)
areas[i].append(t.areaPerUnitT())
l = np.array([[np.median(x), iqr(x)] for x in areas])
N = len(modes)
fig, ax = plt.subplots()
ind = np.arange(N) # the x locations for the groups
width = 0.35 # the width of the bars
fig, ax = plt.subplots()
ax.boxplot(areas)
# add some text for labels, title and axes ticks
ax.set_ylabel('Area covered per unit time (m^2/second)')
ax.set_xticks(ind + 1)
ax.set_xticklabels(modes, rotation='vertical')
plt.xlabel('Mode of Transport')
plt.show()
return l
def batchTraj2Image(df, root, size, side):
'''
Takes the inventory dataframe (containing N trajectories) and the root path of data
Returns:
- An (N,size,size) array of image data
- An array of Transportation Mode labels of length N
'''
df = df.loc[df['Point Count'] > 20][
df['Label-state'] != 'Unlabelled'].copy()
df = df.reset_index(drop=True)
imgs = np.zeros((len(df), size, size))
labels = []
for i in tqdm(range(len(df))):
t = trajectory(root + '/' + df.loc[i]['Path'])
imgs[i] = makeMat(t, windowRange(imgCentre(t), side), size)
labels.append(t.mode)
return imgs, np.array(labels)
def find_lowHurst(root):
'''
Returns a list with the paths of trajectories with small hurst exponents
'''
df = idf[idf['Length'] > 20][idf['Point Count'] > 20][
idf['Duration'] > 0.5][idf['Duration'] < 60].sample(1000)
df = df.reset_index(drop=True)
odd_hurst = []
bar = progressbar.ProgressBar(max_value=len(df))
for index, row in df.iterrows():
bar.update(index)
t = trajectory(root + '/' + row['Path'])
t.removeNoise()
if t.trashy:
continue
if len(t.points) < 41:
continue
if t.hurst() < 0.5:
odd_hurst.append(t.path)
return odd_hurst