please define the input data variables

[khavernathy]

Hi, I can see the input data is a list of dictionaries that contain

• "mmsi": self-explanatory
• "traj": N x 6 vector

The N appears to be an arbitrary number of AIS transmissions for the history of an MMSI. What are the 6 float values specifically?

Could you please explain or point to the part of the paper that describes these inputs?

For example

>>> f = pickle.load(open(".../TrAISformer/data/ct_dma/ct_dma_test.pkl", "rb"))
>>> f[0]["traj"][0]
array([8.64908000e-02, 8.63617407e-01, 3.53333333e-01, 1.55555556e-01,
       1.55318019e+09, 2.09155000e+08])

[khavernathy]

It appears to be

lat
lon
SOG
COG
start time
end time

is that right? For a given MMSI, all of the "start time" and "end time" are the same, though.

[khavernathy]

I see also that

lat, lon, sog, and cod have been standardized, i.e. range = [0,1).

Could you specify how this was done?

[dnguyengithub]

Hello,

1. The format of traj is [lat, log, sog, cog, unix_timestamp, mmsi]

2. To scale the data, just do x_scaled = (x - x_min)/(x_max - x_min)

[khavernathy]

Many thanks. What is the x_max for sog? 30 knots? Sorry to bother so much, I just want to get it right. @dnguyengithub

[dnguyengithub]

Yes 30 knots

[khavernathy]

@dnguyengithub is the maximum course 72? (or 360?) for purposes of scaling 0->1. The code in config_trAISformer.py seems to suggest that it's 72.

    if dataset_name == "ct_dma": #==============================

        # When mode == "grad" or "pos_grad", sog and cog are actually dlat and 
        # dlon    
        lat_size = 250
        lon_size = 270
        sog_size = 30
        cog_size = 72

[sontung2310]

@khavernathy Hi, do you get the answer for the maximum course? Does it 72 or 360? Thank you!