Thanks for this excellent contribution, Haider @hwasiti I plan to try some of these as I progress further into modeling.
@sgugger Some of my friends asked me to make pull request to include those pretrained models into fastai.
Any chance that you will add them to fastai in the future? or can I make a pull request but then where exactly in the library should I make the changes to add them?
i have learnt that if we pass lin_ftrs= then soon after adaptive create cnn will create Linear layers with no of Inputs as we put in the list… so we dont need toput our own custom head ,if our requirement is just to change the last Fc layer features with output as no of classses for various trained models…
@jeremy I had some trouble with your macro in the basic worksheet, with the last line in your onestep subroutine. It kept giving me a 400 error. I fixed it with the following changes. First, I changed the subroutine to accept an integer as an argument:
Sub onestep(ByVal i as Integer)
Then, I changed the last line to
Range("msestart").Offset(i, 0).PasteSpecial Paste:=xlPasteValues
And calling the subroutine is then simply
For i = 1 To 5: onestep (i): Next
Thanks for the post! The link you provided in the second update is a really nice reference. I was able to get everything working on my end, though I had a one thing not working as expected:
get_groups() Fastai methods? If so I think they’ve been replaced by something else or just removed completely. I can’t find anything on the docs. There’s other stuff in the github repo that isn’t updated with the most recent Fastai version, but those two seemed really nice to learn how these models are structured/setting layer groups.
Thanks again for the post!
@jeremy @sgugger How to use Collaborative filtering for the non-rating dataset.e.g. product recommendation based on transaction history. I used TuriCreate API.But, how to implement product recommendation problem based on the transaction in fastai?
Secondly, do fastai support various type of recommendations as present in TuriCreate API?
That’s pretty cool. Do you know of anyone using this in practice?
Thank you @jeremy for the excellent lecture. I have been thinking much about collaborative filtering and I have a few questions to ask.
In the example given, the item list in the collaborative filtering dataset does not necessarily need to be movies. It can be anything with a rating to it. Potentially one can put demographic or other user data in there (e.g. if the person is male, then the entry male can be made in the movie column with a rating of 5 to capture this) and the model can learn how these data can influence the user embedding. This may avoid requiring a separate tabular model to use demographic data for the cold start problem. I wonder if anyone has tried this approach?
The more interesting idea is to use collaborative filtering in medical diagnostics. Most patients only have a few diseases, and most diseases only affect a small proportion of patients, so this is not dissimilar to the movie recommendation problem. Potentially, if there is a dataset of a large number of patients with their diagnostic coding, then a collaborative filtering system can figure out if what diseases a patient may be most susceptible to given their previous diagnoses.
Collaborative filtering can also be used to impute missing values. This is again of importance in medicine as most patients will not have had all the tests / investigations that is available. Potentially as more information is known about a patient, a system can impute and predict test results that have not been performed.
Does anyone know if a collaborative filtering approach has been used in medical research / diagnostics?