search

songlab-cal / tape-neurips2019

Tasks Assessing Protein Embeddings (TAPE), a set of five biologically relevant semi-supervised learning tasks spread across different domains of protein biology. (DEPRECATED)
https://arxiv.org/abs/1906.08230
MIT License
118 stars 34 forks source link

Pretrained supervised task-specific weights for fluorescence task? #6

Open spark157 opened 5 years ago

spark157 commented 5 years ago

Hello,

I'm trying to run the fluorescence task with fully pretrained model weights (ie. for both the unsupervised pretraining and weights for the supervised task as well). I have downloaded the pretrained UniRep model but I'm thinking this does not include the model weights for the supervised task (fluorescence). Are these available or am I confusing something?

In the README.md|Loading a Model there is a bit about loading supervised task-specific weights but it wasn't clear to me.

Thanks for any guidance you have.

Scott

thomas-a-neil commented 5 years ago

Hi Scott,

You can find the fluorescence task model weights here, where *_pretrain* indicates the model was first pretrained on language modeling, then fine-tuned to fluorescence. You should be able to load them with load_task_from=</path/to/supervised_weights.h5> note that this is different than the load_from argument

http://s3.amazonaws.com/proteindata/fluorescence_weights/bepler_pretrain_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/lstm_pretrain_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/lstm_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/one_hot_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/resnet_pretrain_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/resnet_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/transformer_pretrain_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/transformer_weights.h5 http://s3.amazonaws.com/proteindata/fluorescence_weights/unirep_pretrain_weights.h5

I will work on making the other pretrained task models available, and making the instructions on how to access them clearer in the README.

jannisborn commented 2 years ago

Hi @captaincapsaicin and @spark157 it seems that that the links have expired. Do you still have those weights available? A quick reply would be greatly appreciated. I'm working toward a deadline and would like to include this as a comparison for a paper

thomas-a-neil commented 2 years ago

Hi @jannisborn, we recently migrated our S3 hosting. The updated URLs for fluorescence task model weights are here.

http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/bepler_pretrain_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/lstm_pretrain_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/lstm_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/one_hot_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/resnet_pretrain_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/resnet_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/transformer_pretrain_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/transformer_weights.h5 http://s3.amazonaws.com/songlabdata/proteindata/fluorescence_weights/unirep_pretrain_weights.h5

Keep in mind that this repo is an archived version from our neurips 2019 submission, while https://github.com/songlab-cal/tape is under development

jannisborn commented 2 years ago

Thanks for the quick reply @captaincapsaicin 👍🏼
But unfortunately the buckets dont seem to be publicly exposed. I get "AccessDenied" errors for all files :/

thomas-a-neil commented 2 years ago

Thanks for catching that, they should be publicly exposed now.

jannisborn commented 2 years ago

Indeed, works now, thanks a lot!