search

yaodi833 / trajectory2vec

code for "Trajectory clustering via deep representation learning"
180 stars 55 forks source link

Usage with a custom dataset #3

Open rbhatia46 opened 4 years ago

rbhatia46 commented 4 years ago

Hi, thanks for the amazing implementation. I had one question, what does the simulated trajectories denote here? I can see 3 features in each of the trajectories in sim_trajectories, what do these represent, I have a dataset containing latitude, longitude and timestamp and want to use this algorithm for that.

Can you please let me know about what do those 3 features represent(assuming they are timestamp,x,y), also what changes do I need to make to use this for GPS data, so that I can use it, any help is appreciated :)

yaodi833 commented 4 years ago

Thanks for your interest. For using this repo on real trajectory containing latitude, longitude and timestamp), two necessary operations are feature extraction and normalization. Before feeding to the model, all feature values in the time window should be normalized to (0,1).

rbhatia46 commented 4 years ago

@yaodi833 thanks for the reverting back, and I have the data from GPS, so I have it in CRS : EPSG:4326, do I need to convert them back to a CRS suitable for cartesian system, because I see you have used euclidean distance and other formulas for cartesian system, could you suggest me a suitable CRS? Thanks

yaodi833 commented 4 years ago

Sorry, I am not very familiar with the data format in GIS and can not give you further suggestion. Actually, the inputs of the neural network model are numerical features containing the speed, acceleration, and etc. One thing you need to do is to figure out the way to obtain these features with your own dataset.
More details about the feature can be found in the related paper.