![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6231775%2F5f8cbb64604c44e8019f1043d96b14e6%2FScreenshot%202022-08-18%20144054.png?generation=1660814723722264&alt=media\")
no matter what kind of loss function I use I get loss == non .. is there any one who can help me out ??
", "messages": [ { "id": "1904552", "postDate": "08/18/2022 09:27:26", "content": "
no matter what kind of loss function I use I get loss == non .. is there any one who can help me out ??
", "rawMarkdown": "\n\n\nno matter what kind of loss function I use I get loss == non .. is there any one who can help me out ??", "votes": null }, { "id": "1904604", "postDate": "08/18/2022 10:34:05", "content": "You are evaluating first epoch, wait after first epoch and tell if there is still nan value.
", "rawMarkdown": "You are evaluating first epoch, wait after first epoch and tell if there is still nan value.", "votes": null }, { "id": "1904634", "postDate": "08/18/2022 11:20:11", "content": "Fp16?
\nLower learning rate
@sakvaua sir lr=0.001 , what do you suggest ??
", "rawMarkdown": "sakvaua sir lr=0.001 , what do you suggest ??", "votes": null }, { "id": "1904765", "postDate": "08/18/2022 13:16:57", "content": "@urosjarc sir
\n ![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6231775%2Fd8db77af148c1b8f103ddaac3582d799%2FScreenshot%202022-08-18%20191334.png?generation=1660828508639575&alt=media\")
\nššš
Its interesting that accuracy is evaluating but loss is not, can you post the code so that I can check what you are doing please.
", "rawMarkdown": "Its interesting that accuracy is evaluating but loss is not, can you post the code so that I can check what you are doing please.", "votes": null }, { "id": "1905063", "postDate": "08/18/2022 18:01:28", "content": "I don't know how much you will understand . Please ignore my noob coding. you can clone it from here .
\nhttps://github.com/AmzadHossainrafis/project2
", "rawMarkdown": "I don't know how much you will understand . Please ignore my noob coding. you can clone it from here .\n\n [https://github.com/AmzadHossainrafis/project2](url)", "votes": null }, { "id": "1905112", "postDate": "08/18/2022 18:34:46", "content": "Oh you have full project on github, joust like me,
\ncan you create minimal working example I can help you then, but I will not debug whole project.
can i have your mail . kaggle is not allowing to share github
", "rawMarkdown": "can i have your mail . kaggle is not allowing to share github", "votes": null }, { "id": "1905361", "postDate": "08/19/2022 02:30:10", "content": "Put it on gist: https://gist.github.com/ so that every body can see and help you.
", "rawMarkdown": "Put it on gist: https://gist.github.com/ so that every body can see and help you.", "votes": null }, { "id": "1907832", "postDate": "08/21/2022 05:56:03", "content": "Hi @amzadhossainrafi, make sure you've used a non-linear activation function like sigmoid or softmax in your final convolution layer.
", "rawMarkdown": "Hi @amzadhossainrafi, make sure you've used a non-linear activation function like sigmoid or softmax in your final convolution layer.", "votes": null }, { "id": "1908438", "postDate": "08/21/2022 17:13:24", "content": "thq sir for your advice . But I already solve this problem . I was facing this prb becoz of my mask loading process ..
", "rawMarkdown": "thq sir for your advice . But I already solve this problem . I was facing this prb becoz of my mask loading process ..", "votes": null }, { "id": "1909052", "postDate": "08/22/2022 09:09:50", "content": "You can try official loss functions, such as nn.CrossEntropyLoss(). If this does not solve your problem, you can try print the value of your prediction and the value of ground truth, and check what is wrong.
", "rawMarkdown": "You can try official loss functions, such as nn.CrossEntropyLoss(). If this does not solve your problem, you can try print the value of your prediction and the value of ground truth, and check what is wrong.", "votes": null } ], "comments": [ { "id": 1904604, "author_name": "urosjarc", "author_url": "", "post_date": "08/18/2022 10:34:05", "content": "You are evaluating first epoch, wait after first epoch and tell if there is still nan value.
", "votes": null, "replies": [] }, { "id": 1904634, "author_name": "sakvaua", "author_url": "", "post_date": "08/18/2022 11:20:11", "content": "Fp16?
\nLower learning rate
@sakvaua sir lr=0.001 , what do you suggest ??
", "votes": null, "replies": [] }, { "id": 1904765, "author_name": "amzadhossainrafi", "author_url": "", "post_date": "08/18/2022 13:16:57", "content": "@urosjarc sir
\n
\nššš
Its interesting that accuracy is evaluating but loss is not, can you post the code so that I can check what you are doing please.
", "votes": null, "replies": [] }, { "id": 1905063, "author_name": "amzadhossainrafi", "author_url": "", "post_date": "08/18/2022 18:01:28", "content": "I don't know how much you will understand . Please ignore my noob coding. you can clone it from here .
\nhttps://github.com/AmzadHossainrafis/project2
", "votes": null, "replies": [] }, { "id": 1905112, "author_name": "urosjarc", "author_url": "", "post_date": "08/18/2022 18:34:46", "content": "Oh you have full project on github, joust like me,
\ncan you create minimal working example I can help you then, but I will not debug whole project.
can i have your mail . kaggle is not allowing to share github
", "votes": null, "replies": [] }, { "id": 1905361, "author_name": "urosjarc", "author_url": "", "post_date": "08/19/2022 02:30:10", "content": "Put it on gist: https://gist.github.com/ so that every body can see and help you.
", "votes": null, "replies": [] } ] }, { "id": 1907832, "author_name": "maunilshah", "author_url": "", "post_date": "08/21/2022 05:56:03", "content": "Hi @amzadhossainrafi, make sure you've used a non-linear activation function like sigmoid or softmax in your final convolution layer.
", "votes": null, "replies": [ { "id": 1908438, "author_name": "amzadhossainrafi", "author_url": "", "post_date": "08/21/2022 17:13:24", "content": "thq sir for your advice . But I already solve this problem . I was facing this prb becoz of my mask loading process ..
", "votes": null, "replies": [] } ] }, { "id": 1909052, "author_name": "gray98", "author_url": "", "post_date": "08/22/2022 09:09:50", "content": "You can try official loss functions, such as nn.CrossEntropyLoss(). If this does not solve your problem, you can try print the value of your prediction and the value of ground truth, and check what is wrong.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1904552": "\n\n\nno matter what kind of loss function I use I get loss == non .. is there any one who can help me out ??", "1904604": "You are evaluating first epoch, wait after first epoch and tell if there is still nan value.", "1904634": "Fp16?\nLower learning rate", "1904764": "sakvaua sir lr=0.001 , what do you suggest ??", "1904765": "urosjarc sir \n  \nššš", "1905027": "Its interesting that accuracy is evaluating but loss is not, can you post the code so that I can check what you are doing please.", "1905063": "I don't know how much you will understand . Please ignore my noob coding. you can clone it from here .\n\n [https://github.com/AmzadHossainrafis/project2](url)", "1905112": "Oh you have full project on github, joust like me,\ncan you create minimal working example I can help you then, but I will not debug whole project.", "1905315": "can i have your mail . kaggle is not allowing to share github", "1905361": "Put it on gist: https://gist.github.com/ so that every body can see and help you.", "1907832": "Hi @amzadhossainrafi, make sure you've used a non-linear activation function like sigmoid or softmax in your final convolution layer.", "1908438": "thq sir for your advice . But I already solve this problem . I was facing this prb becoz of my mask loading process ..", "1909052": "You can try official loss functions, such as nn.CrossEntropyLoss(). If this does not solve your problem, you can try print the value of your prediction and the value of ground truth, and check what is wrong." }, "source": "meta" }