dbvis-ukon / movekit Goto Github PK

Compute the distance matrix between the entities for each time step
In addition compute the mean distance between entities

• Compute the medoid for each time frame

• Create a example notebook for the clustering with detailed description how to use and what the results are

There is some indexing issue in the function which produces the same result for each tie frame
This maybe solves the issue
for time in data_time.keys(): final_matrix[time] = pd.DataFrame(distance_matrix(data_time[time].loc[:, ['x', 'y']].values, data_time[time].loc[:, ['x', 'y']]), index = data_time[time].loc[:, 'animal_id'].values, columns = data_time[time].loc[:, 'animal_id'].values)

Improve styling of all code pages, specifically considering indentation, max line length, blank lines, single vs double quotes etc. Use PEP8 as foundation.

Movement-based kernel density estimates

Read and parse excel file as input source

Change the imports for the cluster example notebook

needed for distence_centroid function right afterwards

Enable the computation of DTW for the trajectories - so a distance matrix between all trajectories
This should be computationally expensive

Extend the demo by depicting the computed features for each animal in a chart
Also display the resulting Json in a readable way in the notebook

• divide by zero encountered in double_scalars direction = math.degrees(math.atan((y2 - y1) / (x2 - x1)))
• RuntimeWarning: invalid value encountered in double_scalar direction = math.degrees(math.atan((y2 - y1) / (x2 - x1)))

Use scipy distance functions

Add fuzzy temporal segmentation - e,g, if the dataset is segmented into 10 min intervals. Add a window of 2 minutes the overlaps on either side of the segments.

Add to preprocessing

function to calculate network and hierarchy of the group

function to get distance from each animal to centroid at each time

Traveled distance between the first point in the animal trajectory and the last point - sum of traveled distance per animal

Add general description of the package and its functionalities, list of dependencies, citation with zenodo doi, development statement to the readme

• Measure the exploration of the environment (e.g. in percentage or a heatmap)

Replace subsets (segments) of a animals movement based on some indices e.g. time

Compute the distance to the centroid of the swarm

Combine mediod and centroid computation in one method

Acceleration can be currently also minus add another feature which is called pos_acceleration and has only the positive acceleration

• Compute for speed, acceleration, metric distance etc. the mean, median, variance, IQR
• Use the tsfresh library for this

• Compute the distance to a object - input is a coordinate and then the library computes the distance to this object for each time frame

Create a documentation and a online page with Sphinx

• temporal autocorrelation of speed and heading to determine how quickly fish change in their movements
• use tsfresh

Remove dead code - or add comments to the code snippet to with #dead to indicate that the code is currently not used

• Use a outlier detection algorithm
• Use a parameter 'e' to determine the distance to some features which specifies if a mover is an outlier

Document each function with doc-strings that work with SPHINX

Apply a filter on the temporal dimension to filter specific time intervals

• Convert the measures e.g. (x,y) from pixel space to mm or m etc.

Implement the optional parameter in CSV and excel

Plot the missing values of an animal id against the time. Add to preprocess

Compute the minimum convex polygon to compute the animal homrange using a minimum convex polygon. Compute the mimumum convex polygon for the whole animal group and each animal individually.

• provide the option to calculate this from x,y data (‘heading’) or to actually use a data column where angle is provided from the tracking directly (‘orientation’)

• compute turning speed

• Rename files to just io, feature_extraction etc. Similar to the files in src/movekit
• Combine test files into a single file for each py file in src

Compute the maximum distance between two consecutive points in time (largest metric distance between two frames)

Split the trajectory of a single animal into several intervals (segments) according to some specific criterion.

Splitting may be interesting for example to detect different properties in time intervals. E.g. split into segments of 1 minute

Incorporate time series clustering for absolute features
*Write a wrapper for the following method Link

• Input again a location (x,y) and a minimum distance e then the duration for each animal at that location is computed

• Implement a ST DBSCAN

• proportion of time moving/not-moving based on the stopped feature

Resample the movement data of each animal - by downsampling at fixed time intervals.
This can be done systematically (every minute pick a element from an animal) or random downsampling (the same random time intervals for each animal).

When installing in new environment, the following error always arises:

ERROR: tsfresh 0.12.0 has requirement pandas<=0.23.4,>=0.20.3, but you'll have pandas 0.24.2 which is incompatible.

Also there are issues with the "init.py" file as it seems not able to load the relative files, like the csv module in the io folder:

>>> import movekit
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/Jolle/.virtualenvs/jolpy3/lib/python2.7/site-packages/movekit/__init__.py", line 1, in <module>
    from .io.csv import *
ImportError: No module named io.csv

Write pytest for every submodule - see pytest

Compute the voronoi diagram for each time step and compute also the area for each cell over time

Document all ts-fresh parameters for the mkit.ts_feature(data_features, 'autocorrelation')

See example demo for this

Multiprocess the compute_absolute_features method.