I would really like more explanation of the factorized top k Brute Force and SCANN methods. The written docs don't say much, and I can sort of understand based on context what it is doing but would love a more detailed explanation on those two.
at 9:31 (ruclips.net/video/jz0-satrmrA/видео.html) what is the colon in __ini__ function for? For example the part with "self.task: tf.keras.layers.Layer = task"
This is just a type annotation. It means that he's passing the type of each argument to easy understanding the matter of each argument. It's optional, so you could write directly "self.task = task" instead of "self.task: tf.keras.layers.Layer = task"
I noticed in this video, they are using the userId as an input to the model to generate the vector embeddings for what movies that user likes. What happens when we add more users with new userIds? Do we have to retrain the model every time?
Yes. One other solution if you don't want to retrain the model every time is to create an embedding for ids that are not in the training set. This embedding will be used foe new users. It is similar to out of vocab tokens in nlp.
@@MrAmgadHasan are you here bro? I keep getting 'error:Cannot batch tensors with different shapes in component 0. First element had shape [1] and element 1 had shape [2]. [Op:IteratorGetNext]' whenever I try to this at 5:35 --> movie_titles = movies.batch(1_000).map(lambda x: x['movie_title']) user_ids = ratings.batch(1_000_000).map(lambda x: x["user_id"]) unique_movie_titles = np.unique(np.concatenate(list(movie_titles))) unique_user_ids = np.unique(np.concatenate(list(user_ids))) unique_movie_titles[:4] why? help me out here please
why use id random number as embedding? why using identity matrix as label? pls i need further explanation about. tf.eye as label for loss function will be helpful thanks!
I would really like more explanation of the factorized top k Brute Force and SCANN methods. The written docs don't say much, and I can sort of understand based on context what it is doing but would love a more detailed explanation on those two.
Interesting. I can prob. make a video to better explain approximate neighbor search (which is what ScaNN does) so that it makes more sense.
@@windmaple why use id as embeddings instead of genre and gender? It make more sense?
at 9:31 (ruclips.net/video/jz0-satrmrA/видео.html) what is the colon in __ini__ function for? For example the part with "self.task: tf.keras.layers.Layer = task"
This is just a type annotation. It means that he's passing the type of each argument to easy understanding the matter of each argument. It's optional, so you could write directly "self.task = task" instead of "self.task: tf.keras.layers.Layer = task"
@@lebesgue-integral ol
O
I noticed in this video, they are using the userId as an input to the model to generate the vector embeddings for what movies that user likes. What happens when we add more users with new userIds? Do we have to retrain the model every time?
Yes. One other solution if you don't want to retrain the model every time is to create an embedding for ids that are not in the training set. This embedding will be used foe new users. It is similar to out of vocab tokens in nlp.
@@MrAmgadHasan are you here bro? I keep getting 'error:Cannot batch tensors with different shapes in component 0. First element had shape [1] and element 1 had shape [2]. [Op:IteratorGetNext]' whenever I try to this at 5:35 --> movie_titles = movies.batch(1_000).map(lambda x: x['movie_title'])
user_ids = ratings.batch(1_000_000).map(lambda x: x["user_id"])
unique_movie_titles = np.unique(np.concatenate(list(movie_titles)))
unique_user_ids = np.unique(np.concatenate(list(user_ids)))
unique_movie_titles[:4]
why? help me out here please
Can I a create recommender system model for my streaming mobile app ? If yes, how?
Start coding.
Tensorflow Lite will help you here.
This is tough
this is so tough bro
Please make Firebase Extension for this recommendation system 🤣
wow just wow
Thank you 🙌
Not Recommnded videos to watch,
why use id random number as embedding? why using identity matrix as label? pls i need further explanation about. tf.eye as label for loss function will be helpful thanks!
Thank you !