I am facing memory error on the training dataset.
\n\n1) Even 14 GB RAM CPU is not sufficient for non keras models. I tried implementing sentence embedding from weighting word embeddings.
\n\n2) Even 6 GB GPU is not sufficient for training keras models. Facing Resource Exhausted issue. Any help would be much appreciated.
\n\nI tried even batch_norm for speeding up the process. As it is kernel based competition any help would be highly appreciated.
", "messages": [ { "id": 418096, "postDate": "2018-11-09T09:30:00.993Z", "content": "I am facing memory error on the training dataset.
\n\n1) Even 14 GB RAM CPU is not sufficient for non keras models. I tried implementing sentence embedding from weighting word embeddings.
\n\n2) Even 6 GB GPU is not sufficient for training keras models. Facing Resource Exhausted issue. Any help would be much appreciated.
\n\nI tried even batch_norm for speeding up the process. As it is kernel based competition any help would be highly appreciated.
", "rawMarkdown": "I am facing memory error on the training dataset.\n\n1) Even 14 GB RAM CPU is not sufficient for non keras models. I tried implementing sentence embedding from weighting word embeddings.\n\n2) Even 6 GB GPU is not sufficient for training keras models. Facing Resource Exhausted issue. Any help would be much appreciated.\n\nI tried even batch_norm for speeding up the process. As it is kernel based competition any help would be highly appreciated.", "votes": 1 }, { "id": 418128, "postDate": "2018-11-09T10:41:10.337Z", "content": "Are you me? I'm trying to implement Arora et al's Simple but Tough-to-Beat Baseline and facing the same issue.
\n\nHowever I do think that with a few more tweaks I can make it work. Sure, limited resources suck but this is still better than being in an arm race with some guy who can ensemble 500 models. It's a fair game.
", "rawMarkdown": "Are you me? I'm trying to implement Arora et al's Simple but Tough-to-Beat Baseline and facing the same issue.\n\nHowever I do think that with a few more tweaks I can make it work. Sure, limited resources suck but this is still better than being in an arm race with some guy who can ensemble 500 models. It's a fair game.", "votes": 2, "replies": [ { "id": 418133, "postDate": "2018-11-09T10:58:28.663Z", "content": "Hey, Yes that was in my mind too, implement SIF for sentence embeddings, I am facing an issue while doing some descriptive statistics on word embedding for sentence embeddings. Yes, I have to optimize the code further. The issue is kernel getting restarted.
", "rawMarkdown": "Hey, Yes that was in my mind too, implement SIF for sentence embeddings, I am facing an issue while doing some descriptive statistics on word embedding for sentence embeddings. Yes, I have to optimize the code further. The issue is kernel getting restarted." }, { "id": 418164, "postDate": "2018-11-09T11:48:49.870Z", "rawMarkdown": "", "isDeleted": true }, { "id": 418319, "postDate": "2018-11-09T16:34:18.523Z", "content": "I released my SIF kernel here https://www.kaggle.com/suicaokhoailang/gpu-free-mlp-baseline-with-sif
\n\nLet me know what you think.
", "rawMarkdown": "I released my SIF kernel here https://www.kaggle.com/suicaokhoailang/gpu-free-mlp-baseline-with-sif\n\nLet me know what you think." } ] }, { "id": 456705, "postDate": "2019-01-16T10:42:30.697Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 418128, "author_name": "Khoi Nguyen", "author_url": "", "post_date": "2018-11-09T10:41:10.337000", "content": "Are you me? I'm trying to implement Arora et al's Simple but Tough-to-Beat Baseline and facing the same issue.
\n\nHowever I do think that with a few more tweaks I can make it work. Sure, limited resources suck but this is still better than being in an arm race with some guy who can ensemble 500 models. It's a fair game.
", "votes": 2, "replies": [ { "id": 418133, "author_name": "Karthik Chowdary Tsaliki", "author_url": "", "post_date": "2018-11-09T10:58:28.663000", "content": "Hey, Yes that was in my mind too, implement SIF for sentence embeddings, I am facing an issue while doing some descriptive statistics on word embedding for sentence embeddings. Yes, I have to optimize the code further. The issue is kernel getting restarted.
", "votes": 0, "replies": [] }, { "id": 418164, "author_name": "", "author_url": "", "post_date": "2018-11-09T11:48:49.870000", "content": "", "votes": 0, "replies": [] }, { "id": 418319, "author_name": "Khoi Nguyen", "author_url": "", "post_date": "2018-11-09T16:34:18.523000", "content": "I released my SIF kernel here https://www.kaggle.com/suicaokhoailang/gpu-free-mlp-baseline-with-sif
\n\nLet me know what you think.
", "votes": 0, "replies": [] } ] }, { "id": 456705, "author_name": "", "author_url": "", "post_date": "2019-01-16T10:42:30.697000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "418096": "I am facing memory error on the training dataset.\n\n1) Even 14 GB RAM CPU is not sufficient for non keras models. I tried implementing sentence embedding from weighting word embeddings.\n\n2) Even 6 GB GPU is not sufficient for training keras models. Facing Resource Exhausted issue. Any help would be much appreciated.\n\nI tried even batch_norm for speeding up the process. As it is kernel based competition any help would be highly appreciated.", "418128": "Are you me? I'm trying to implement Arora et al's Simple but Tough-to-Beat Baseline and facing the same issue.\n\nHowever I do think that with a few more tweaks I can make it work. Sure, limited resources suck but this is still better than being in an arm race with some guy who can ensemble 500 models. It's a fair game.", "456705": "" } }