Add Tutorial for MSTUMP/MSTUMPED

[seanlaw]

The original MSTUMP paper is here.

The data and code for mSTAMP can be found here.

For the 3-dimensional toy dataset shown in Figure 2, the data can be retrieved by:

#!/usr/bin/env python

from scipy.io import loadmat
import ssl
import io
import urllib
import pandas as pd

context = ssl.SSLContext()

url = "https://github.com/mcyeh/mstamp/blob/master/Python/toy_data.mat?raw=true"

raw_bytes = urllib.request.urlopen(url, context=context).read()
data = io.BytesIO(raw_bytes)
mat = loadmat(data)

mdata = mat['data']
mdtype = mdata.dtype

df = pd.DataFrame(mdata, dtype=mdtype, columns=['dim1', 'dim2', 'dim3'])
print(df)

[seanlaw]

An initial Jupyter notebook has already been started here.

[seanlaw]

Example using NFL player tracking data might be interesting

Or even player movement tracking in NBA basketball is even better! Here's an example and another paper that has some nice ideas.

[seanlaw]

Consider adding an example of how to use the "Inclusion" multi-dimensional motif search

[seanlaw]

None of this really matters until we are able to compute a "Matrix Profile Subspace" and "MDL" for choosing k

[seanlaw]

The boxing data can be read using:

!pip install transforms3d
!git clone https://github.com/CalciferZh/AMCParser
import sys
sys.path.append('AMCParser')
import amc_parser as amc

joints = amc.parse_asf("13.asf")
motions = amc.parse_amc("13_17.amc")

xs, ys, zs = [], [], []
for frame in range(len(motions)):
    joints['root'].set_motion(motions[frame])
    c_joints = joints['root'].to_dict()
    for joint in c_joints.values():
#         print(joint.name, joint.coordinate[0, 0], joint.coordinate[1, 0], joint.coordinate[2, 0])
        xs.append(joint.coordinate[0, 0])
        ys.append(joint.coordinate[1, 0])
        zs.append(joint.coordinate[2, 0])

frame_rate = 120
print(len(motions) / frame_rate)  # Total number of seconds is ~40.3 s

Note that the paper mentions a 10 millisecond window size in Figure 13 but this is not possible since it would be less than one frame.