Hi All!
\nI am really new to Recommender systems and have recently been on a β΄ voyage to find some resources. Here are the ones I liked:
\nI have been recently exploring the NVIDIA Merlin ecosystem.
\nThey have a fantastic set of libraries and frameworks that let allow you to use GPUs for accelerating your workflows. I found their example repository to be very detailed.
\nI also had the oppurtunity to interview some grandmasters from the recent ACM RecSys winning team:
\nACM Recsys competitions might have some great resources/papers from winners, I'll try to update this thread as I continue finding them.
\nSee you all on the LB! π΅
", "messages": [ { "id": "1680446", "postDate": "02/07/2022 19:56:44", "content": "Hi All!
\nI am really new to Recommender systems and have recently been on a β΄ voyage to find some resources. Here are the ones I liked:
\nI have been recently exploring the NVIDIA Merlin ecosystem.
\nThey have a fantastic set of libraries and frameworks that let allow you to use GPUs for accelerating your workflows. I found their example repository to be very detailed.
\nI also had the oppurtunity to interview some grandmasters from the recent ACM RecSys winning team:
\nACM Recsys competitions might have some great resources/papers from winners, I'll try to update this thread as I continue finding them.
\nSee you all on the LB! π΅
", "rawMarkdown": "Hi All! \n\nI am really new to Recommender systems and have recently been on a β΄ voyage to find some resources. Here are the ones I liked:\n\n## Open Source: NVIDIA Merlin | GPU + Recommenders = π\n\nI have been recently exploring the [NVIDIA Merlin](https://github.com/NVIDIA-Merlin/Merlin) ecosystem. \n\nThey have a fantastic set of libraries and frameworks that let allow you to use GPUs for accelerating your workflows. I found their [example repository](https://github.com/NVIDIA-Merlin/Merlin/tree/main/examples) to be very detailed. \n\n\n## Videos/Interviews/Blogposts\n\nI also had the oppurtunity to interview some grandmasters from the recent ACM RecSys winning team:\n\n- ACM RecSys 2021 winning solution [Video](https://www.youtube.com/watch?v=W3aWEXqIkWk), [Audio](https://anchor.fm/chaitimedatascience/episodes/ACM-RecSys-Winning-Solution-Benedikt-Schifferer--Bo-Liu--Chris-Deotte--Even-Oldridge-136-e156mlo/a-a6868jn)\n- [How to Build a Winning Deep Learning Recommender System by NVIDIA](https://www.youtube.com/watch?v=bHuww-l_Sq0&list=PL5B692fm6--uXbxtmPJz5nu3Xmc1JUm3F)\n\nACM Recsys competitions might have some great resources/papers from winners, I'll try to update this thread as I continue finding them. \n\nSee you all on the LB! π΅", "votes": null }, { "id": "1681337", "postDate": "02/08/2022 12:24:24", "content": "Last I checked the merlin package, it didn't support new data / novel schemasβ¦ Has that changed?
", "rawMarkdown": "Last I checked the merlin package, it didn't support new data / novel schemas... Has that changed?", "votes": null }, { "id": "1681437", "postDate": "02/08/2022 13:28:01", "content": "I'll be playing around this week, will let you know if it has changed.
\nI'm really new to Recommender systems so my plan is to learn via this competition and I was hoping to publish a starter pack while using Merlin
", "rawMarkdown": "I'll be playing around this week, will let you know if it has changed. \n\nI'm really new to Recommender systems so my plan is to learn via this competition and I was hoping to publish a starter pack while using Merlin", "votes": null }, { "id": "1684598", "postDate": "02/10/2022 15:42:53", "content": "Hello, I am working at NVIDIA in the Merlin team. I am exciting to hear, that you checked out our packages. What new data / novel schemas do you miss?
\nWe provide many examples in our GitHub Repository:
\nBasic examples: https://github.com/NVIDIA-Merlin/NVTabular/tree/main/examples
\nTransformer Architectures for sequential RecSys: https://github.com/NVIDIA-Merlin/Transformers4Rec/tree/main/examples
\nModel library: https://github.com/NVIDIA-Merlin/models
The last one is active in development and we will provide many examples, soon
", "rawMarkdown": "Hello, I am working at NVIDIA in the Merlin team. I am exciting to hear, that you checked out our packages. What new data / novel schemas do you miss?\n\nWe provide many examples in our GitHub Repository:\nBasic examples: https://github.com/NVIDIA-Merlin/NVTabular/tree/main/examples\nTransformer Architectures for sequential RecSys: https://github.com/NVIDIA-Merlin/Transformers4Rec/tree/main/examples\nModel library: https://github.com/NVIDIA-Merlin/models\n\nThe last one is active in development and we will provide many examples, soon", "votes": null }, { "id": "1684773", "postDate": "02/10/2022 17:53:44", "content": "I am also from the NVIDIA Merlin team. We have just shared a GPU-accelerated EDA for this comp based on our RecSys Dataset Profiler template notebook.
", "rawMarkdown": "I am also from the NVIDIA Merlin team. We have just shared a [GPU-accelerated EDA](https://www.kaggle.com/gspmoreira/h-m-recsys-dataset-profiler-eda-gpu-acc) for this comp based on our **RecSys Dataset Profiler** template notebook.", "votes": null }, { "id": "1686461", "postDate": "02/12/2022 05:06:03", "content": "Great to see you here Gabriel!
\nQuick Question: Did you get a chance to benchmark GPU accelerated EDA against using vanilla pandas?
\nBased on my readups about the Merlin team's achievements-I understand it really shines with larger datasets. I'm curious about the performance here.
\nMaybe its a homework for me to try π
", "rawMarkdown": "Great to see you here Gabriel! \n\nQuick Question: Did you get a chance to benchmark GPU accelerated EDA against using vanilla pandas? \n\nBased on my readups about the Merlin team's achievements-I understand it really shines with larger datasets. I'm curious about the performance here. \n\nMaybe its a homework for me to try π ", "votes": null }, { "id": "1687072", "postDate": "02/12/2022 15:48:18", "content": "ohohoo back to competing on Kaggle β¦ good choice @init27 π
", "rawMarkdown": "ohohoo back to competing on Kaggle ... good choice @init27 π", "votes": null }, { "id": "1687124", "postDate": "02/12/2022 16:21:06", "content": "Our chat inspired me π
", "rawMarkdown": "Our chat inspired me π", "votes": null }, { "id": "1687168", "postDate": "02/12/2022 16:46:51", "content": "Glad to hear that. Enjoy!
", "rawMarkdown": "Glad to hear that. Enjoy!", "votes": null }, { "id": "1702579", "postDate": "02/23/2022 18:10:16", "content": "Nice find @init27
\nThanks a lot and have a good competition!
Hi @init27 . You are right, GPU-accelerated data frame operations really shines in particular on larger datasets. I have not created a Pandas version of that notebook for comparison. We just wanted to share that template notebook as it is a good resource for quickly collecting useful statistics on recsys datasets, which can help doing some design decisions on how to better model the recommendation problem.
", "rawMarkdown": "Hi @init27 . You are right, GPU-accelerated data frame operations really shines in particular on larger datasets. I have not created a Pandas version of that notebook for comparison. We just wanted to share that template notebook as it is a good resource for quickly collecting useful statistics on recsys datasets, which can help doing some design decisions on how to better model the recommendation problem.", "votes": null }, { "id": "1821268", "postDate": "06/15/2022 11:35:26", "content": "I feel quite hard to get any useful resources in Kaggle for the most classical recommender system algorithm - The latent Factor Model
\nSo, I tried to build a simple Recommender System from scratch by this algorithm (mainly NumPy used, no other 3rd party package).
You can check it if you are interested: https://www.kaggle.com/code/ddatad/recommender-system-latent-factor-model
", "rawMarkdown": "I feel quite hard to get any useful resources in Kaggle for the most classical recommender system algorithm - The latent Factor Model\nSo, I tried to build a simple Recommender System from scratch by this algorithm (mainly NumPy used, no other 3rd party package).\n\nYou can check it if you are interested: https://www.kaggle.com/code/ddatad/recommender-system-latent-factor-model", "votes": null } ], "comments": [ { "id": 1681337, "author_name": "danofer", "author_url": "", "post_date": "02/08/2022 12:24:24", "content": "Last I checked the merlin package, it didn't support new data / novel schemas⦠Has that changed?
", "votes": null, "replies": [ { "id": 1681437, "author_name": "init27", "author_url": "", "post_date": "02/08/2022 13:28:01", "content": "I'll be playing around this week, will let you know if it has changed.
\nI'm really new to Recommender systems so my plan is to learn via this competition and I was hoping to publish a starter pack while using Merlin
", "votes": null, "replies": [] }, { "id": 1684598, "author_name": "benediktschifferer", "author_url": "", "post_date": "02/10/2022 15:42:53", "content": "Hello, I am working at NVIDIA in the Merlin team. I am exciting to hear, that you checked out our packages. What new data / novel schemas do you miss?
\nWe provide many examples in our GitHub Repository:
\nBasic examples: https://github.com/NVIDIA-Merlin/NVTabular/tree/main/examples
\nTransformer Architectures for sequential RecSys: https://github.com/NVIDIA-Merlin/Transformers4Rec/tree/main/examples
\nModel library: https://github.com/NVIDIA-Merlin/models
The last one is active in development and we will provide many examples, soon
", "votes": null, "replies": [] } ] }, { "id": 1684773, "author_name": "gspmoreira", "author_url": "", "post_date": "02/10/2022 17:53:44", "content": "I am also from the NVIDIA Merlin team. We have just shared a GPU-accelerated EDA for this comp based on our RecSys Dataset Profiler template notebook.
", "votes": null, "replies": [ { "id": 1686461, "author_name": "init27", "author_url": "", "post_date": "02/12/2022 05:06:03", "content": "Great to see you here Gabriel!
\nQuick Question: Did you get a chance to benchmark GPU accelerated EDA against using vanilla pandas?
\nBased on my readups about the Merlin team's achievements-I understand it really shines with larger datasets. I'm curious about the performance here.
\nMaybe its a homework for me to try π
", "votes": null, "replies": [] }, { "id": 1702795, "author_name": "gspmoreira", "author_url": "", "post_date": "02/23/2022 23:24:35", "content": "Hi @init27 . You are right, GPU-accelerated data frame operations really shines in particular on larger datasets. I have not created a Pandas version of that notebook for comparison. We just wanted to share that template notebook as it is a good resource for quickly collecting useful statistics on recsys datasets, which can help doing some design decisions on how to better model the recommendation problem.
", "votes": null, "replies": [] } ] }, { "id": 1687072, "author_name": "crodoc", "author_url": "", "post_date": "02/12/2022 15:48:18", "content": "ohohoo back to competing on Kaggle β¦ good choice @init27 π
", "votes": null, "replies": [ { "id": 1687124, "author_name": "init27", "author_url": "", "post_date": "02/12/2022 16:21:06", "content": "Our chat inspired me π
", "votes": null, "replies": [] }, { "id": 1687168, "author_name": "crodoc", "author_url": "", "post_date": "02/12/2022 16:46:51", "content": "Glad to hear that. Enjoy!
", "votes": null, "replies": [] } ] }, { "id": 1702579, "author_name": "datascientistfp", "author_url": "", "post_date": "02/23/2022 18:10:16", "content": "Nice find @init27
\nThanks a lot and have a good competition!
I feel quite hard to get any useful resources in Kaggle for the most classical recommender system algorithm - The latent Factor Model
\nSo, I tried to build a simple Recommender System from scratch by this algorithm (mainly NumPy used, no other 3rd party package).
You can check it if you are interested: https://www.kaggle.com/code/ddatad/recommender-system-latent-factor-model
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1680446": "Hi All! \n\nI am really new to Recommender systems and have recently been on a β΄ voyage to find some resources. Here are the ones I liked:\n\n## Open Source: NVIDIA Merlin | GPU + Recommenders = π\n\nI have been recently exploring the [NVIDIA Merlin](https://github.com/NVIDIA-Merlin/Merlin) ecosystem. \n\nThey have a fantastic set of libraries and frameworks that let allow you to use GPUs for accelerating your workflows. I found their [example repository](https://github.com/NVIDIA-Merlin/Merlin/tree/main/examples) to be very detailed. \n\n\n## Videos/Interviews/Blogposts\n\nI also had the oppurtunity to interview some grandmasters from the recent ACM RecSys winning team:\n\n- ACM RecSys 2021 winning solution [Video](https://www.youtube.com/watch?v=W3aWEXqIkWk), [Audio](https://anchor.fm/chaitimedatascience/episodes/ACM-RecSys-Winning-Solution-Benedikt-Schifferer--Bo-Liu--Chris-Deotte--Even-Oldridge-136-e156mlo/a-a6868jn)\n- [How to Build a Winning Deep Learning Recommender System by NVIDIA](https://www.youtube.com/watch?v=bHuww-l_Sq0&list=PL5B692fm6--uXbxtmPJz5nu3Xmc1JUm3F)\n\nACM Recsys competitions might have some great resources/papers from winners, I'll try to update this thread as I continue finding them. \n\nSee you all on the LB! π΅", "1681337": "Last I checked the merlin package, it didn't support new data / novel schemas... Has that changed?", "1681437": "I'll be playing around this week, will let you know if it has changed. \n\nI'm really new to Recommender systems so my plan is to learn via this competition and I was hoping to publish a starter pack while using Merlin", "1684598": "Hello, I am working at NVIDIA in the Merlin team. I am exciting to hear, that you checked out our packages. What new data / novel schemas do you miss?\n\nWe provide many examples in our GitHub Repository:\nBasic examples: https://github.com/NVIDIA-Merlin/NVTabular/tree/main/examples\nTransformer Architectures for sequential RecSys: https://github.com/NVIDIA-Merlin/Transformers4Rec/tree/main/examples\nModel library: https://github.com/NVIDIA-Merlin/models\n\nThe last one is active in development and we will provide many examples, soon", "1684773": "I am also from the NVIDIA Merlin team. We have just shared a [GPU-accelerated EDA](https://www.kaggle.com/gspmoreira/h-m-recsys-dataset-profiler-eda-gpu-acc) for this comp based on our **RecSys Dataset Profiler** template notebook.", "1686461": "Great to see you here Gabriel! \n\nQuick Question: Did you get a chance to benchmark GPU accelerated EDA against using vanilla pandas? \n\nBased on my readups about the Merlin team's achievements-I understand it really shines with larger datasets. I'm curious about the performance here. \n\nMaybe its a homework for me to try π ", "1687072": "ohohoo back to competing on Kaggle ... good choice @init27 π", "1687124": "Our chat inspired me π", "1687168": "Glad to hear that. Enjoy!", "1702579": "Nice find @init27 \nThanks a lot and have a good competition!", "1702795": "Hi @init27 . You are right, GPU-accelerated data frame operations really shines in particular on larger datasets. I have not created a Pandas version of that notebook for comparison. We just wanted to share that template notebook as it is a good resource for quickly collecting useful statistics on recsys datasets, which can help doing some design decisions on how to better model the recommendation problem.", "1821268": "I feel quite hard to get any useful resources in Kaggle for the most classical recommender system algorithm - The latent Factor Model\nSo, I tried to build a simple Recommender System from scratch by this algorithm (mainly NumPy used, no other 3rd party package).\n\nYou can check it if you are interested: https://www.kaggle.com/code/ddatad/recommender-system-latent-factor-model" }, "source": "meta" }