model weights and inference

[kvnsng]

Hello - thanks for this very interesting paper and work!

I was just wondering if you had model weights and inference code we can try. Thank you!

[Mr81wx]

loading_ckpt https://mail.google.com/mail/u/0?ui=2&ik=96449053f0&attid=0.1&permmsgid=msg-a:r-7239896480587092773&view=att&disp=safe&realattid=f_m1anbuxg0 (17 K) https://mail.google.com/mail/u/0?ui=2&ik=96449053f0&attid=0.1&permmsgid=msg-a:r-7239896480587092773&view=att&disp=safe&realattid=f_m1anbuxg0Hi Kevin,

Sorry for the late reply.

I attached the checkpoints for our model. You can load the parameters under Pytorch Lightning's mode.

However, there are strict limitations on the input format of the model, you can refer to the attached file for the create_batch_class function, this is used for classification finetuning.

Best

Xing Checkpoints_C.ckpt https://drive.google.com/open?id=1qdcXb9CRJZgniNVaViiZaQXNiWxpiTnQ

Kevin Song @.***> 于2024年9月12日周四 20:22写道：

Hello - thanks for this very interesting paper and work!

I was just wondering if you had model weights and inference code we can try. Thank you!

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/BBSM4RGFNFSIN75LKI4ESRDZWHLU5AVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43ASLTON2WKOZSGUZDGMBWGMYTQMA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[kvnsng]

Thank you for the access! I don't think I have the file with create_batch_class function. Would you be able to add that to the github repo? If not, could you send it to my email? songkevinc@gmail.com

[Mr81wx]

try this link https://colab.research.google.com/drive/1YCGmMn5CBC0j8d5sj45Y53FcAvxvoJWE?usp=drive_link

Kevin Song @.***> 于2024年9月20日周五 15:13写道：

Thank you for the access! I don't think I have the file with create_batch_class function. Would you be able to add that to the github repo? If not, could you send it to my email? @.***

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363710210, or unsubscribe https://github.com/notifications/unsubscribe-auth/BBSM4RCAQHWOVIX56NNLNN3ZXQNPXAVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTG4YTAMRRGA . You are receiving this because you commented.Message ID: @.***>

[kvnsng]

That works. Thank you and I'll go ahead and close the issue!

[Mr81wx]

Great!

We have updated the new model extensions for defense players. Please let me know if you need them.

We're also looking for some soccer and football data to see this framework's performance. If you have similar resources and ideas, we can work together!

2024年9月20日 下午3:17，Kevin Song @.***> 写道：

That works. Thank you and I'll go ahead and close the issue!

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363718978, or unsubscribe https://github.com/notifications/unsubscribe-auth/BBSM4RHMXBZD4Q2C6AQNRNDZXQOALAVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTG4YTQOJXHA. You are receiving this because you commented.

[kvnsng]

Sounds good. thank you!

[kvnsng]

Another quick question on the play embeddings... I noticed the output from the encoder is in the shape of [6,121,256]. When you create play embeddings, do you just flatten this tensor? I couldn't find the method in the paper

[Mr81wx]

average pooling

Kevin Song @.***>于2024年9月20日 周五16:27写道：

Another quick question on the play embeddings... I noticed the output from the encoder is in the shape of [6,121,256]. When you create play embeddings, do you just flatten this tensor? I couldn't find the method in the paper

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363866801, or unsubscribe https://github.com/notifications/unsubscribe-auth/BBSM4RASYRCACZ5KX3IKGILZXQWEZAVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTHA3DMOBQGE . You are receiving this because you commented.Message ID: @.***>

[kvnsng]

Gotcha. Do you mind sharing the parameters you used for average pooling as well? kernel_size, stride?

[Mr81wx]

we just take mean() at the 1st and 2nd dimensions，kernel size 5*121. There may be better parameters here.

Kevin Song @.***>于2024年9月20日 周五16:42写道：

Gotcha. Do you mind sharing the parameters you used for average pooling as well? kernel_size, stride?

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363898700, or unsubscribe https://github.com/notifications/unsubscribe-auth/BBSM4RF4U523C7DVC44TR3DZXQX47AVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTHA4TQNZQGA . You are receiving this because you commented.Message ID: @.***>

[kvnsng]

Got it. Thank you!. Just to confirm, this results in embedding size of [256]? I just did torch.mean(vector, dim=[0,1])

On Fri, Sep 20, 2024 at 10:50 AM Mr81wx @.***> wrote:

we just take mean() at the 1st and 2nd dimensions，kernel size 5*121. There may be better parameters here.

Kevin Song @.***>于2024年9月20日 周五16:42写道：

Gotcha. Do you mind sharing the parameters you used for average pooling as well? kernel_size, stride?

— Reply to this email directly, view it on GitHub < https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363898700>,

or unsubscribe < https://github.com/notifications/unsubscribe-auth/BBSM4RF4U523C7DVC44TR3DZXQX47AVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTHA4TQNZQGA>

. You are receiving this because you commented.Message ID: @.***>

— Reply to this email directly, view it on GitHub https://github.com/Mr81wx/HoopTransformer/issues/1#issuecomment-2363914572, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFZB6ZG27YN4QDZRKRSQPXLZXQY2NAVCNFSM6AAAAABOD2O2GWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRTHEYTINJXGI . You are receiving this because you modified the open/close state.Message ID: @.***>