The accuracy curve and the loss curve that i am getting is not smooth at all. Since I am not able to share the image because they have dissabled the sharing from local system. But the general trend of the accuracy curve is in the upwards, and for the loss curve, it is downwards, but there are many spikes in between.
\nI tried with low learning rates, different batch sizes, and some other things, but the curves that i am getting is not at all smooth in any of the cases. What else should I try to smoothen the curve.
The accuracy curve and the loss curve that i am getting is not smooth at all. Since I am not able to share the image because they have dissabled the sharing from local system. But the general trend of the accuracy curve is in the upwards, and for the loss curve, it is downwards, but there are many spikes in between.
\nI tried with low learning rates, different batch sizes, and some other things, but the curves that i am getting is not at all smooth in any of the cases. What else should I try to smoothen the curve.
I guess it is under fitting, try different model or a larger one with more iterations. If they are occasional spikes then no need to worry, but if the graph is random then there is a problem.
", "rawMarkdown": "I guess it is under fitting, try different model or a larger one with more iterations. If they are occasional spikes then no need to worry, but if the graph is random then there is a problem.", "votes": null }, { "id": "1300032", "postDate": "05/10/2021 08:14:20", "content": "I got the similar results but only on the validation dataset. The loss and F1 curve of training is smooth but on validation set, they fluctuates severely.
", "rawMarkdown": "I got the similar results but only on the validation dataset. The loss and F1 curve of training is smooth but on validation set, they fluctuates severely.", "votes": null }, { "id": "1300050", "postDate": "05/10/2021 08:34:06", "content": "I guess I am getting this for the training curve because in the fit method while I am training the model, I am not following steps_per_epoch = (no_of_training_data/batch_size)
", "rawMarkdown": "I guess I am getting this for the training curve because in the fit method while I am training the model, I am not following steps_per_epoch = (no_of_training_data/batch_size)", "votes": null }, { "id": "1300059", "postDate": "05/10/2021 08:37:52", "content": "Yeah you are right, it is underfitting, and i figured out the reason, it is because I am not following steps_per_epoch = (no_of_training_data/batch_size), I am taking pretty low value for the steps_per_epoch, because of which it is not training on the all the batches for a single epoch.
", "rawMarkdown": "Yeah you are right, it is underfitting, and i figured out the reason, it is because I am not following steps_per_epoch = (no_of_training_data/batch_size), I am taking pretty low value for the steps_per_epoch, because of which it is not training on the all the batches for a single epoch.", "votes": null }, { "id": "1303440", "postDate": "05/12/2021 04:28:44", "content": "Interestingly ,I was worried about it too ! I used a 90:10 , train-validation split , then two , the validation loss as well as accuracy was fluctuating but when I submitted the notebook , it gave promising results. Though still working on solving this problem and still on the mission of finding out the exact reasons why it's fluctuating but giving a promising public score
", "rawMarkdown": "Interestingly ,I was worried about it too ! I used a 90:10 , train-validation split , then two , the validation loss as well as accuracy was fluctuating but when I submitted the notebook , it gave promising results. Though still working on solving this problem and still on the mission of finding out the exact reasons why it's fluctuating but giving a promising public score", "votes": null }, { "id": "1306403", "postDate": "05/13/2021 18:33:20", "content": "How much accuracy are you getting on your validation set? I am getting 75%, and then it is fluctuating between 75-80%.
", "rawMarkdown": "How much accuracy are you getting on your validation set? I am getting 75%, and then it is fluctuating between 75-80%.", "votes": null }, { "id": "1306493", "postDate": "05/13/2021 20:01:11", "content": "After much Hyperparameter tuning ,it is reaching it is fluctuating between 88-91 . You can find one of my published notebooks here : https://www.kaggle.com/amartyabhattacharya/plant-pathology-2021-eda-resnet-50v2-model
", "rawMarkdown": "After much Hyperparameter tuning ,it is reaching it is fluctuating between 88-91 . You can find one of my published notebooks here : https://www.kaggle.com/amartyabhattacharya/plant-pathology-2021-eda-resnet-50v2-model", "votes": null } ], "comments": [ { "id": 1297103, "author_name": "jimitshah777", "author_url": "", "post_date": "05/07/2021 18:29:13", "content": "I guess it is under fitting, try different model or a larger one with more iterations. If they are occasional spikes then no need to worry, but if the graph is random then there is a problem.
", "votes": null, "replies": [ { "id": 1300059, "author_name": "jl18pg052", "author_url": "", "post_date": "05/10/2021 08:37:52", "content": "Yeah you are right, it is underfitting, and i figured out the reason, it is because I am not following steps_per_epoch = (no_of_training_data/batch_size), I am taking pretty low value for the steps_per_epoch, because of which it is not training on the all the batches for a single epoch.
", "votes": null, "replies": [] } ] }, { "id": 1300032, "author_name": "crissallan", "author_url": "", "post_date": "05/10/2021 08:14:20", "content": "I got the similar results but only on the validation dataset. The loss and F1 curve of training is smooth but on validation set, they fluctuates severely.
", "votes": null, "replies": [ { "id": 1300050, "author_name": "jl18pg052", "author_url": "", "post_date": "05/10/2021 08:34:06", "content": "I guess I am getting this for the training curve because in the fit method while I am training the model, I am not following steps_per_epoch = (no_of_training_data/batch_size)
", "votes": null, "replies": [] } ] }, { "id": 1303440, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/12/2021 04:28:44", "content": "Interestingly ,I was worried about it too ! I used a 90:10 , train-validation split , then two , the validation loss as well as accuracy was fluctuating but when I submitted the notebook , it gave promising results. Though still working on solving this problem and still on the mission of finding out the exact reasons why it's fluctuating but giving a promising public score
", "votes": null, "replies": [ { "id": 1306403, "author_name": "jl18pg052", "author_url": "", "post_date": "05/13/2021 18:33:20", "content": "How much accuracy are you getting on your validation set? I am getting 75%, and then it is fluctuating between 75-80%.
", "votes": null, "replies": [] }, { "id": 1306493, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/13/2021 20:01:11", "content": "After much Hyperparameter tuning ,it is reaching it is fluctuating between 88-91 . You can find one of my published notebooks here : https://www.kaggle.com/amartyabhattacharya/plant-pathology-2021-eda-resnet-50v2-model
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1293844": "The accuracy curve and the loss curve that i am getting is not smooth at all. Since I am not able to share the image because they have dissabled the sharing from local system. But the general trend of the accuracy curve is in the upwards, and for the loss curve, it is downwards, but there are many spikes in between. \nI tried with low learning rates, different batch sizes, and some other things, but the curves that i am getting is not at all smooth in any of the cases. What else should I try to smoothen the curve.", "1297103": "I guess it is under fitting, try different model or a larger one with more iterations. If they are occasional spikes then no need to worry, but if the graph is random then there is a problem.", "1300032": "I got the similar results but only on the validation dataset. The loss and F1 curve of training is smooth but on validation set, they fluctuates severely.", "1300050": "I guess I am getting this for the training curve because in the fit method while I am training the model, I am not following steps_per_epoch = (no_of_training_data/batch_size)", "1300059": "Yeah you are right, it is underfitting, and i figured out the reason, it is because I am not following steps_per_epoch = (no_of_training_data/batch_size), I am taking pretty low value for the steps_per_epoch, because of which it is not training on the all the batches for a single epoch.", "1303440": "Interestingly ,I was worried about it too ! I used a 90:10 , train-validation split , then two , the validation loss as well as accuracy was fluctuating but when I submitted the notebook , it gave promising results. Though still working on solving this problem and still on the mission of finding out the exact reasons why it's fluctuating but giving a promising public score", "1306403": "How much accuracy are you getting on your validation set? I am getting 75%, and then it is fluctuating between 75-80%.", "1306493": "After much Hyperparameter tuning ,it is reaching it is fluctuating between 88-91 . You can find one of my published notebooks here : https://www.kaggle.com/amartyabhattacharya/plant-pathology-2021-eda-resnet-50v2-model" }, "source": "meta" }