Hi, I am very interested in TFMs and have been following your posts on the subject! They have been very clear and interesting. I still do not understand how a TFM would work in deployment. I assume I call the .fit with few labelled data I have and then give the model some batches of data I want to make predictions on. Am I right? Another question I have is when -in deployment- I obtain more labelled data and I want to refine the model. What should I do in that case?
Thanks! API-wise usage is the same as with "classic" machine learning models. Use .fit() and .predict(). If the training data remains the same, you can use the same object that was fitted for making predictions. If the training data changes, you call .fit() again. The case you describe with additional labelled data would mean concatenating the new labelled data to the exististing training data. This becomes the new context. And you use the concatenated data in .fit().
Hi, I am very interested in TFMs and have been following your posts on the subject! They have been very clear and interesting. I still do not understand how a TFM would work in deployment. I assume I call the .fit with few labelled data I have and then give the model some batches of data I want to make predictions on. Am I right? Another question I have is when -in deployment- I obtain more labelled data and I want to refine the model. What should I do in that case?
Thanks! API-wise usage is the same as with "classic" machine learning models. Use .fit() and .predict(). If the training data remains the same, you can use the same object that was fitted for making predictions. If the training data changes, you call .fit() again. The case you describe with additional labelled data would mean concatenating the new labelled data to the exististing training data. This becomes the new context. And you use the concatenated data in .fit().