ValueError while training the embedding layer

[ronger-git]

An error occurred while training the embedding layer：ValueError：Dimensions must be equal,but are 702 and 602 for 'sub' with input shapes:[?,?,702],[1,1,602].

As a beginner, I can’t solve this problem, hope your reply！

[duonginspace]

Hi,
This error happened because you had set the wrong value for the four-hot vector (see create_AIS_dataset in datasets.py).

If you are a beginner, just pull the repository (to update your code), run the code with the default values of the hyperparameters to see how it works.

Good luck!

[ronger-git]

Thanks for your advice, I will continue to study your code and papers.

[Tata17]

Hi @dnguyengithub I hope this message finds you well.

I already ran your dataset and it worked well after changing things here and there.

For my own dataset, I ran csvtopkl.py, dataset_preprocessing.py, and calculate_AIS_mean (in that order).

Now I am getting this same error. I looked at dataset.py > create_AIS_dataset, and I still don't see what parameters to change that could help me. Could I bother you to elaborate a little bit more on that (on the 4-hot vector values)?

As an aside note:

• I didn't have the ROT or Nav_STT columns, so I used the following:
  LAT, LON, SOG, COG, HEADING, TIMESTAMP, MMSI, SHIPTYPE = list(range(8))
  I looked for each file that needed this line and changed it to match my data.
  Not sure if that is causing the dimension mismatch, although I don't think it matters for the 4-hot vectors.

Thank you for your time

[Tata17]

@ronger-git I just realized that you might had solved this issue. Do you have any suggestions?

Best

[duonginspace]

@Tata17 sorry for the late reply.
Your issue relates to the dimension of the 4-hot vector. Did you choose the right values for the 4-hot vector?

[nate-ellis]

@dnguyengithub I'm interested in this as well. Just finished running the embedding layer and about to preprocess my data. Where exactly do we choose the 4-hot vector in the scripts? As in, with the create_AIS_dataset function is this the setting:
"msg,lat_bins = 300, lon_bins = 300, sog_bins = 30 ,cog_bins = 72"

Also, I have everything but ROT. How much will that impact the output?

Thanks in advance for the assist.

[duonginspace]

@Tata17 @nate-ellis Hello, sorry for taking so long to answer your question, I was busy preparing my thesis defense.
The dimensions of the 4-hot vector are hard-coded (yes, it's bad) in datasets.py, you should change those values.

@nate-ellis You dont need ROT for this version (ROT is used for another version of GeoTrackNet). Only LAT, LON, SOG, COG, MMSI and TIMESTAMP are required in this version.

I will update a better version when I have time.