spotlight
spotlight copied to clipboard
maciejkula
Deep recommender models using PyTorch.
I do not think the present docstring for the ZeroEmbedding makes sense because you are not scaling anything, its just initialized to zero.
While training implicit sequence models, we use losses like hinge, bpr and pointwise. These losses don't maximize directly the dot product, so why do we use it while predicting?
I'm trying to download spotlight on my Mac with Python 3.7 and I'm getting this error: ``` bash-3.2# conda install -c maciejkula -c pytorch spotlight=0.1.5 Collecting package metadata (current_repodata.json): done...
My questions might seem dumb, but I was trying to understand the project. As seen in the documentation, various parameters go into the model. I was thinking if there is...
Hello, I just notice that you start count id from 1, which lead to one more dimension abuse in both user and item. For example, the number of user and...
may you clarify if this code https://maciejkula.github.io/spotlight/datasets/synthetic.html can be used to generate not sequential data meaning what parameter to set to get not sequential data but something like usual data...
may you share some links to synthetic unbalanced data generating for classification when your code is for recommendation system data https://maciejkula.github.io/spotlight/datasets/synthetic.html meaning close to real data - with mix of...
I looked through the sequence model, it looks like the model is trying to learn a user representation from the given sequence alone: `user_representation, _ = self._net.user_representation( sequence_var )` It...
Hello, I notice that the Interactions class can export data to Sparse Matrix format, but I'm wondering if there's any way to train a model using an already created Sparse...
I am wondering if Spotlight includes the iMF model proposed in _'Collaborative Filtering for Implicit Feedback Datasets'_ by **Yifan Hu** (not BPR)?
