Hi. Firstly, thank you for this work and sharing the code.
I was able to check out summer and go through the setup procedure and re-train with only some minor code updates (see various one liner PRs for the latest dependencies).
Q: To run inference on a screenplay I need to ensure I use the same Universal Sentence Encoder embeddings as I understand it - are you using the official tensor flow implementation of USE to extract embeddings, or is there an equivalent PyTorch implementation you have used?
I ask because even a re- implementation of USE would have different embeddings and would require training from scratch?
And at that point we would need to re-extract embeddings from the TRIPOD dataset and re-train from scratch?
Just triple checking my assumptions.
Thank you again.
Hi. Firstly, thank you for this work and sharing the code.
I was able to check out summer and go through the setup procedure and re-train with only some minor code updates (see various one liner PRs for the latest dependencies).
Q: To run inference on a screenplay I need to ensure I use the same
Universal Sentence Encoderembeddings as I understand it - are you using the official tensor flow implementation ofUSEto extract embeddings, or is there an equivalent PyTorch implementation you have used?I ask because even a re- implementation of USE would have different embeddings and would require training from scratch?
And at that point we would need to re-extract embeddings from the TRIPOD dataset and re-train from scratch?
Just triple checking my assumptions.
Thank you again.