In [1]:
import os
import seaborn as sns
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# import custom functions
os.chdir('..')
from FLOCK_GPS import DataLoading, Preprocessing, DirectionalCorrelation
Load data
In [2]:
# Initialize path to data (UTM-converted datasets)
data_dir = os.getcwd() + '\\SampleData'
# Load datasets
raw_datasets = DataLoading.load_data(data_dir)
# Re-shape datasets
datasets = DataLoading.pivot_datsets(raw_datasets)
Get the interpolated data
In [3]:
# get interpolated datasets
interp_datasets = Preprocessing.interpolate_datasets(datasets, threshold = 0.99)
Get movement periods
In [4]:
# get slices for movement periods and break times
movements_bySquad, rests_bySquad, all_stops = Preprocessing.get_slices_byArea(interp_datasets, plot=False)
Extracting movement periods
100%|██████████| 5/5 [09:12<00:00, 110.45s/it]
Get an just one example squad's data
In [5]:
movements = movements_bySquad[1]
Get the smoothed data
In [6]:
smooth_movements = Preprocessing.spline_smoothing(movements, s=3e1, UTM=True)
Get leadership metrics
In [7]:
names = smooth_movements[0].UTM_x.columns
time_delay_dfs, HCS_ratio_dfs, graphs = DirectionalCorrelation.get_directional_corr(smooth_movements, names=names, UTM=True, window_length = 9)
100%|██████████| 7/7 [05:57<00:00, 51.07s/it]
This gives us a 'leadership score' from the average time-delay for each individual
The more positive the time-delay, the more directional influence that individual has
In [8]:
time_delay_dfs[0].mean()
Out[8]:
Member 1 (Squad 2) 0.028342 Member 2 (Squad 2) 0.086318 Member 3 (Squad 2) -0.098910 Member 4 (Squad 2) -0.046654 Member 5 (Squad 2) 0.072385 Member 6 (Squad 2) -0.068123 Member 7 (Squad 2) -0.091568 Member 8 (Squad 2) 0.128615 dtype: float64
We also get a metric for Highly Correlated Segments (HCS)
This is the ratio of time an individual is highly correlated with other individuals while they are within proximity
In [9]:
HCS_ratio_dfs[0].mean()
Out[9]:
Member 1 (Squad 2) 0.788374 Member 2 (Squad 2) 0.875661 Member 3 (Squad 2) 0.847850 Member 4 (Squad 2) 0.866153 Member 5 (Squad 2) 0.832689 Member 6 (Squad 2) 0.829647 Member 7 (Squad 2) 0.811997 Member 8 (Squad 2) 0.796939 dtype: float64
Plot leadership graphs for each movement period
The Y axis location of nodes represents the time-delay (leadership score) value, while the X axis is arbitrary
The edges point from leader to follower with a specific directional correlation time-delay for that pair
The Quality of the heirarchy is considered the ratio of arrows pointing downards (from high leadership score to low leadership score)
The number of loops should be close to zero
We mark the individual who is Squad leader with green and the platoon leaders with blue
In [11]:
sq_name = smooth_movements[0].attrs['name'].split('.')[0]
DirectionalCorrelation.leadership_graph_ani(time_delay_dfs, graphs, names, sq_name, show=False)
print('animation saved to ' + os.getcwd() + r'\..\Figures\Pairwise_Leadership_Animation'+sq_name+'.gif')
MovieWriter ffmpeg unavailable; using Pillow instead.
animation saved to c:\Users\James\GitRepos\GPS-ruck\..\Figures\Pairwise_Leadership_AnimationSquad_2.gif
Nodes represent group members
Edges represent the direction of influence as well as the associated time delay
Locaiton on the Y axis represents an individual's average directional correlation tiem delay
Simple simulation
In [3]:
import random
p1 = [[ 0, 0]]
p2 = [[ 8, 0]]
p3 = [[-8, 0]]
steps = 100
for step in range(steps):
p1start = 15
p2start = 10
p3start = 20
if step >= p1start and step < (p1start + np.pi*2*10):
x = step - p1start
p1extra = (np.sin(x/10))/2
else:
p1extra = 0
if step >= p2start and step < (p2start + np.pi*2*10):
x = step - p2start
p2extra = (np.sin(x/10))/2
else:
p2extra = 0
if step >= p3start and step < (p3start + np.pi*2*10):
x = step - p3start
p3extra = (np.sin(x/10))/2
else:
p3extra = 0
p1x = random.randint(-10,10)/100 + p1extra
p2x = random.randint(-10,10)/100 + p2extra
p3x = random.randint(-10,10)/100 + p3extra
p1y = random.randint(-10,10)/100 + 1
p2y = random.randint(-10,10)/100 + 1
p3y = random.randint(-10,10)/100 + 1
p1.append([p1[-1][0] + p1x, p1[-1][1] + p1y])
p2.append([p2[-1][0] + p2x, p2[-1][1] + p2y])
p3.append([p3[-1][0] + p3x, p3[-1][1] + p3y])
Pdf = pd.concat([pd.DataFrame(p, columns=pd.MultiIndex.from_tuples((('UTM_x', 'Member: '+str(s)) , ('UTM_y', 'Member: '+str(s)) ))) for s, p in enumerate ([p3,p1,p2])], axis=1)
colors = {'Member: 0':'r', 'Member: 1': 'g', 'Member: 2': 'b'}
for name in Pdf.UTM_x.columns:
plt.scatter(Pdf.UTM_x[name], Pdf.UTM_y[name], label=name, color=colors[name])
plt.gca().set_aspect('equal')
plt.legend()
sns.despine()
plt.tight_layout()
plt.gca().set_xlim(-10, 60)
plt.gca().set_ylim(-10, 120)
time_delay_dfs, HCS_ratio_dfs, graphs = DirectionalCorrelation.get_directional_corr([Pdf], names=Pdf.UTM_x.columns, UTM=True, window_length = 9)
DirectionalCorrelation.leadership_graph_ani(time_delay_dfs, graphs, names=Pdf.UTM_x.columns, sq_name='test', show=False)
100%|██████████| 1/1 [00:00<00:00, 2.27it/s] MovieWriter ffmpeg unavailable; using Pillow instead.
In [28]:
colors = {'Member: 0':'r', 'Member: 1': 'g', 'Member: 2': 'b'}
for name in Pdf.UTM_x.columns:
plt.scatter(Pdf.UTM_x[name], Pdf.UTM_y[name], label=name, color=colors[name])
plt.axhline(10, color='b', label='Member 2 begins turn')
plt.axhline(15, color='g', label='Member 1 begins turn')
plt.axhline(20, color='r', label='Member 0 begins turn')
plt.gca().set_aspect('equal')
plt.legend()
sns.despine()
plt.tight_layout()
plt.gca().set_xlim(-10, 80)
plt.gca().set_ylim(-10, 120)
Out[28]:
(-10.0, 120.0)
In [31]:
time_delay_dfs[0].mean(), HCS_ratio_dfs[0]
Out[31]:
(Member: 0 -1.306122
Member: 1 -0.016667
Member: 2 1.326531
dtype: float64,
Member: 0 Member: 1 Member: 2
0 0.795775 1.0 0.795775)
In [ ]: