{ "id": 235980, "title": "Number of epochs is not going more than 38.", "url": "/competitions/plant-pathology-2021-fgvc8/discussion/235980", "author_name": "", "post_date": "2021-05-02T07:37:38.558812600Z", "votes": null, "comment_count": 17, "views": 0, "content": "

I tried to train my CNN model for 50 epochs, even for 100 epochs, but it is somehow not going above 38 epochs, and I am not using any early stopping, even though it is not going above 38 eochs, can any one share why is this happening?

", "messages": [ { "id": "1290622", "postDate": "05/02/2021 07:37:38", "content": "

I tried to train my CNN model for 50 epochs, even for 100 epochs, but it is somehow not going above 38 epochs, and I am not using any early stopping, even though it is not going above 38 eochs, can any one share why is this happening?

", "rawMarkdown": "I tried to train my CNN model for 50 epochs, even for 100 epochs, but it is somehow not going above 38 epochs, and I am not using any early stopping, even though it is not going above 38 eochs, can any one share why is this happening?", "votes": null }, { "id": "1290702", "postDate": "05/02/2021 09:53:24", "content": "

Does your notebook run end with error?

", "rawMarkdown": "Does your notebook run end with error?", "votes": null }, { "id": "1290790", "postDate": "05/02/2021 12:13:19", "content": "

No, it doesn't end with any error message

", "rawMarkdown": "No, it doesn't end with any error message", "votes": null }, { "id": "1290795", "postDate": "05/02/2021 12:14:50", "content": "

I think it may be because, since GPU time limit in a single commit is 2 hours, so to not exceed that time limit, this may be happening, but I am not sure.

", "rawMarkdown": "I think it may be because, since GPU time limit in a single commit is 2 hours, so to not exceed that time limit, this may be happening, but I am not sure.", "votes": null }, { "id": "1290816", "postDate": "05/02/2021 12:53:20", "content": "

GPU/TPU have a runtime limit of 9 hours on Kaggle. You have 2 hours GPU limit when you submit your results, so you shouldn't put training into submission. Instead of that train model in separate notebook, and later use obtained weights to do inference.

", "rawMarkdown": "GPU/TPU have a runtime limit of 9 hours on Kaggle. You have 2 hours GPU limit when you submit your results, so you shouldn't put training into submission. Instead of that train model in separate notebook, and later use obtained weights to do inference.", "votes": null }, { "id": "1296420", "postDate": "05/07/2021 08:45:23", "content": "

So should I be saving the model in another notebook, then how can I use that model in another notebook?

", "rawMarkdown": "So should I be saving the model in another notebook, then how can I use that model in another notebook?", "votes": null }, { "id": "1296489", "postDate": "05/07/2021 09:41:21", "content": "

Run your training notebook, and save resulting output as dataset. Then use that dataset in your inference notebook.

", "rawMarkdown": "Run your training notebook, and save resulting output as dataset. Then use that dataset in your inference notebook.", "votes": null }, { "id": "1296836", "postDate": "05/07/2021 14:23:17", "content": "

Sorry for disturbing you again, but once I train a model, and save it, then I don't need to train it again for prediction right? I can use that model directly for prediction right? Just confirming…😃😃

", "rawMarkdown": "Sorry for disturbing you again, but once I train a model, and save it, then I don't need to train it again for prediction right? I can use that model directly for prediction right? Just confirming...😃😃", "votes": null }, { "id": "1296914", "postDate": "05/07/2021 15:09:42", "content": "

No, you train the model one time, and then you use it. That's all.

", "rawMarkdown": "No, you train the model one time, and then you use it. That's all.", "votes": null }, { "id": "1296947", "postDate": "05/07/2021 15:42:15", "content": "

Okay, thank you…

", "rawMarkdown": "Okay, thank you...", "votes": null }, { "id": "1297927", "postDate": "05/08/2021 12:35:59", "content": "

Hey man, I trained a efficientnetB0 model, for 50 epochs, but even that didn't go beyond 38 epochs, so my plan is I will import that trained model, and train it further so that it would learn more. Currently with 38 epochs, it's accuracy is reaching 68%, so if I train for more 30 epochs, then I could get a good accuracy. Is this a good idea?

", "rawMarkdown": "Hey man, I trained a efficientnetB0 model, for 50 epochs, but even that didn't go beyond 38 epochs, so my plan is I will import that trained model, and train it further so that it would learn more. Currently with 38 epochs, it's accuracy is reaching 68%, so if I train for more 30 epochs, then I could get a good accuracy. Is this a good idea?", "votes": null }, { "id": "1297937", "postDate": "05/08/2021 12:41:33", "content": "

What's the runtime of your notebook? You may want to increase batch size.

", "rawMarkdown": "What's the runtime of your notebook? You may want to increase batch size.", "votes": null }, { "id": "1297948", "postDate": "05/08/2021 12:53:39", "content": "

7347 sec. Current batch-size that I am using is 64.

", "rawMarkdown": "7347 sec. Current batch-size that I am using is 64.", "votes": null }, { "id": "1297954", "postDate": "05/08/2021 13:03:43", "content": "

That's strange. Can you share your notebook's run log? The best way to do it is to use pastebin.com (don't paste it to your message).

", "rawMarkdown": "That's strange. Can you share your notebook's run log? The best way to do it is to use pastebin.com (don't paste it to your message).", "votes": null }, { "id": "1298015", "postDate": "05/08/2021 13:51:07", "content": "

Can you tell me why I am getting this error, while loading that trained model in my new notebook
\nValueError: Unknown layer: FixedDropout
\nI think i have used a dropout layer while training it, so should I remove that and train it again, then in the next import this error won't happen. Is it true?

", "rawMarkdown": "Can you tell me why I am getting this error, while loading that trained model in my new notebook\nValueError: Unknown layer: FixedDropout\nI think i have used a dropout layer while training it, so should I remove that and train it again, then in the next import this error won't happen. Is it true?", "votes": null }, { "id": "1298030", "postDate": "05/08/2021 14:01:47", "content": "
!pip install -q ../input/kerasapplications\n!pip install -q ../input/efficientnet-keras-source-code\n
\n

Add this code on top, and don't forget to add corresponding datasets (packages) to your notebook.

", "rawMarkdown": "```\n!pip install -q ../input/kerasapplications\n!pip install -q ../input/efficientnet-keras-source-code\n```\nAdd this code on top, and don't forget to add corresponding datasets (packages) to your notebook.", "votes": null }, { "id": "1298041", "postDate": "05/08/2021 14:12:56", "content": "

There are many kerasapplications in the dataset I am seeing, which one is it, can you share me the links for both please, so that I import the right one. Also I guess, for efficientnet-keras-source-code, we need internet connection, which is not allowed right.

", "rawMarkdown": "There are many kerasapplications in the dataset I am seeing, which one is it, can you share me the links for both please, so that I import the right one. Also I guess, for efficientnet-keras-source-code, we need internet connection, which is not allowed right.", "votes": null }, { "id": "1298054", "postDate": "05/08/2021 14:24:10", "content": "

https://www.kaggle.com/xhlulu/kerasapplications
\nNo, you won't need internet connection. In fact, it's a reason why you use dataset stored on Kaggle.

", "rawMarkdown": "https://www.kaggle.com/xhlulu/kerasapplications\nNo, you won't need internet connection. In fact, it's a reason why you use dataset stored on Kaggle.", "votes": null } ], "comments": [ { "id": 1290702, "author_name": "atamazian", "author_url": "", "post_date": "05/02/2021 09:53:24", "content": "

Does your notebook run end with error?

", "votes": null, "replies": [ { "id": 1290790, "author_name": "jl18pg052", "author_url": "", "post_date": "05/02/2021 12:13:19", "content": "

No, it doesn't end with any error message

", "votes": null, "replies": [] }, { "id": 1290795, "author_name": "jl18pg052", "author_url": "", "post_date": "05/02/2021 12:14:50", "content": "

I think it may be because, since GPU time limit in a single commit is 2 hours, so to not exceed that time limit, this may be happening, but I am not sure.

", "votes": null, "replies": [] }, { "id": 1290816, "author_name": "atamazian", "author_url": "", "post_date": "05/02/2021 12:53:20", "content": "

GPU/TPU have a runtime limit of 9 hours on Kaggle. You have 2 hours GPU limit when you submit your results, so you shouldn't put training into submission. Instead of that train model in separate notebook, and later use obtained weights to do inference.

", "votes": null, "replies": [] }, { "id": 1296420, "author_name": "jl18pg052", "author_url": "", "post_date": "05/07/2021 08:45:23", "content": "

So should I be saving the model in another notebook, then how can I use that model in another notebook?

", "votes": null, "replies": [] }, { "id": 1296489, "author_name": "atamazian", "author_url": "", "post_date": "05/07/2021 09:41:21", "content": "

Run your training notebook, and save resulting output as dataset. Then use that dataset in your inference notebook.

", "votes": null, "replies": [] }, { "id": 1296836, "author_name": "jl18pg052", "author_url": "", "post_date": "05/07/2021 14:23:17", "content": "

Sorry for disturbing you again, but once I train a model, and save it, then I don't need to train it again for prediction right? I can use that model directly for prediction right? Just confirming…😃😃

", "votes": null, "replies": [] }, { "id": 1296914, "author_name": "atamazian", "author_url": "", "post_date": "05/07/2021 15:09:42", "content": "

No, you train the model one time, and then you use it. That's all.

", "votes": null, "replies": [] }, { "id": 1296947, "author_name": "jl18pg052", "author_url": "", "post_date": "05/07/2021 15:42:15", "content": "

Okay, thank you…

", "votes": null, "replies": [] }, { "id": 1297927, "author_name": "jl18pg052", "author_url": "", "post_date": "05/08/2021 12:35:59", "content": "

Hey man, I trained a efficientnetB0 model, for 50 epochs, but even that didn't go beyond 38 epochs, so my plan is I will import that trained model, and train it further so that it would learn more. Currently with 38 epochs, it's accuracy is reaching 68%, so if I train for more 30 epochs, then I could get a good accuracy. Is this a good idea?

", "votes": null, "replies": [] }, { "id": 1297937, "author_name": "atamazian", "author_url": "", "post_date": "05/08/2021 12:41:33", "content": "

What's the runtime of your notebook? You may want to increase batch size.

", "votes": null, "replies": [] }, { "id": 1297948, "author_name": "jl18pg052", "author_url": "", "post_date": "05/08/2021 12:53:39", "content": "

7347 sec. Current batch-size that I am using is 64.

", "votes": null, "replies": [] }, { "id": 1297954, "author_name": "atamazian", "author_url": "", "post_date": "05/08/2021 13:03:43", "content": "

That's strange. Can you share your notebook's run log? The best way to do it is to use pastebin.com (don't paste it to your message).

", "votes": null, "replies": [] }, { "id": 1298015, "author_name": "jl18pg052", "author_url": "", "post_date": "05/08/2021 13:51:07", "content": "

Can you tell me why I am getting this error, while loading that trained model in my new notebook
\nValueError: Unknown layer: FixedDropout
\nI think i have used a dropout layer while training it, so should I remove that and train it again, then in the next import this error won't happen. Is it true?

", "votes": null, "replies": [] }, { "id": 1298030, "author_name": "atamazian", "author_url": "", "post_date": "05/08/2021 14:01:47", "content": "
!pip install -q ../input/kerasapplications\n!pip install -q ../input/efficientnet-keras-source-code\n
\n

Add this code on top, and don't forget to add corresponding datasets (packages) to your notebook.

", "votes": null, "replies": [] }, { "id": 1298041, "author_name": "jl18pg052", "author_url": "", "post_date": "05/08/2021 14:12:56", "content": "

There are many kerasapplications in the dataset I am seeing, which one is it, can you share me the links for both please, so that I import the right one. Also I guess, for efficientnet-keras-source-code, we need internet connection, which is not allowed right.

", "votes": null, "replies": [] }, { "id": 1298054, "author_name": "atamazian", "author_url": "", "post_date": "05/08/2021 14:24:10", "content": "

https://www.kaggle.com/xhlulu/kerasapplications
\nNo, you won't need internet connection. In fact, it's a reason why you use dataset stored on Kaggle.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1290622": "I tried to train my CNN model for 50 epochs, even for 100 epochs, but it is somehow not going above 38 epochs, and I am not using any early stopping, even though it is not going above 38 eochs, can any one share why is this happening?", "1290702": "Does your notebook run end with error?", "1290790": "No, it doesn't end with any error message", "1290795": "I think it may be because, since GPU time limit in a single commit is 2 hours, so to not exceed that time limit, this may be happening, but I am not sure.", "1290816": "GPU/TPU have a runtime limit of 9 hours on Kaggle. You have 2 hours GPU limit when you submit your results, so you shouldn't put training into submission. Instead of that train model in separate notebook, and later use obtained weights to do inference.", "1296420": "So should I be saving the model in another notebook, then how can I use that model in another notebook?", "1296489": "Run your training notebook, and save resulting output as dataset. Then use that dataset in your inference notebook.", "1296836": "Sorry for disturbing you again, but once I train a model, and save it, then I don't need to train it again for prediction right? I can use that model directly for prediction right? Just confirming...😃😃", "1296914": "No, you train the model one time, and then you use it. That's all.", "1296947": "Okay, thank you...", "1297927": "Hey man, I trained a efficientnetB0 model, for 50 epochs, but even that didn't go beyond 38 epochs, so my plan is I will import that trained model, and train it further so that it would learn more. Currently with 38 epochs, it's accuracy is reaching 68%, so if I train for more 30 epochs, then I could get a good accuracy. Is this a good idea?", "1297937": "What's the runtime of your notebook? You may want to increase batch size.", "1297948": "7347 sec. Current batch-size that I am using is 64.", "1297954": "That's strange. Can you share your notebook's run log? The best way to do it is to use pastebin.com (don't paste it to your message).", "1298015": "Can you tell me why I am getting this error, while loading that trained model in my new notebook\nValueError: Unknown layer: FixedDropout\nI think i have used a dropout layer while training it, so should I remove that and train it again, then in the next import this error won't happen. Is it true?", "1298030": "```\n!pip install -q ../input/kerasapplications\n!pip install -q ../input/efficientnet-keras-source-code\n```\nAdd this code on top, and don't forget to add corresponding datasets (packages) to your notebook.", "1298041": "There are many kerasapplications in the dataset I am seeing, which one is it, can you share me the links for both please, so that I import the right one. Also I guess, for efficientnet-keras-source-code, we need internet connection, which is not allowed right.", "1298054": "https://www.kaggle.com/xhlulu/kerasapplications\nNo, you won't need internet connection. In fact, it's a reason why you use dataset stored on Kaggle." }, "source": "meta" }