Thank you for your comment. You can create a TFRecord dataset. Please refer to this tutorial for further detail: www.tensorflow.org/official_models/fine_tuning_bert
Thanks for the explanation. I have two questions 1) is not clear, why in general we need to save data in TFrecords and then use them for modeling ? why not just doing it directly ? 2) I see a lot of tutorial with image analysis - how ever I am interested in tabular data and how one can leverage this functionality ? e.g. how one should do normalization/standardization of tabular data through TF pipeline ? or how to embed various approaches to take care of categorical variables. thanks again.
Thanks for your feedback. TFRecords are specially useful for big data. For ordinary experiments, you do not have to do it. As you mentioned, it can be done by directly using the data. BUT, what if the data is too large and it cannot be stored in the RAM? What if we have input/output bottleneck? This tutorial aims to show a general approach for designing a pipeline. For tabular data, it would be the best to use Pandas and just use the data generator of TensorFlow to process it. Perhaps in the future, you see easier pipelines in TF for non-image data.
آموزش خوبی بود.
دم شما گرم
ما تازه کارا دلمون به همین ویدئو ها خوشه. دست شما درد نکنه سینا جان
همیشه بدرخشی!
Hi can you tell me how to use TFRecords while training models. Would love to connect thanks!
Thank you for your comment. You can create a TFRecord dataset. Please refer to this tutorial for further detail: www.tensorflow.org/official_models/fine_tuning_bert
Thanks for the explanation. I have two questions 1) is not clear, why in general we need to save data in TFrecords and then use them for modeling ? why not just doing it directly ? 2) I see a lot of tutorial with image analysis - how ever I am interested in tabular data and how one can leverage this functionality ? e.g. how one should do normalization/standardization of tabular data through TF pipeline ? or how to embed various approaches to take care of categorical variables. thanks again.
Thanks for your feedback. TFRecords are specially useful for big data. For ordinary experiments, you do not have to do it. As you mentioned, it can be done by directly using the data. BUT, what if the data is too large and it cannot be stored in the RAM? What if we have input/output bottleneck? This tutorial aims to show a general approach for designing a pipeline. For tabular data, it would be the best to use Pandas and just use the data generator of TensorFlow to process it. Perhaps in the future, you see easier pipelines in TF for non-image data.
Great job man, keep making videos :)
Thank you.
Shoutouts man! Well documented and working example!
Thank you!
It's very helpful, thank you very much~ :)
Glad it was helpful!