wandb.log({\"loss\": loss}) in your code that's it :)\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2226462%2F24c5af97669adae32e67a897535b3f31%2FScreenshot%20from%202019-09-28%2012-13-57.png?generation=1569653358570650&alt=media\")
Do you ever wanted to see what is going on inside the code you just committed ? Even more due to recent GPU time limit where you don't want to waste your time in a interactive session. Try Weights and biases (link). its super easy to setup even in kaggle kernel and you can see the log at their website. this way you can terminate your commit early if needed saving time plus it is free. you can keep track of your experiments like hyper parameters etc.
\n\np.s. I am not any way connected to this company.
", "messages": [ { "id": "635733", "postDate": "09/28/2019 05:45:42", "content": "Do you ever wanted to see what is going on inside the code you just committed ? Even more due to recent GPU time limit where you don't want to waste your time in a interactive session. Try Weights and biases (link). its super easy to setup even in kaggle kernel and you can see the log at their website. this way you can terminate your commit early if needed saving time plus it is free. you can keep track of your experiments like hyper parameters etc.
\n\np.s. I am not any way connected to this company.
", "rawMarkdown": "Do you ever wanted to see what is going on inside the code you just committed ? Even more due to recent GPU time limit where you don't want to waste your time in a interactive session. Try Weights and biases [(link)](https://www.wandb.com/). its super easy to setup even in kaggle kernel and you can see the log at their website. this way you can terminate your commit early if needed saving time plus it is free. you can keep track of your experiments like hyper parameters etc.\n\np.s. I am not any way connected to this company.", "votes": null }, { "id": "635760", "postDate": "09/28/2019 06:32:46", "content": "thank you a ton mate for this post,are you using it? is it only for pytorch users? @dhananjay3
", "rawMarkdown": "thank you a ton mate for this post,are you using it? is it only for pytorch users? @dhananjay3", "votes": null }, { "id": "635764", "postDate": "09/28/2019 06:49:48", "content": "1) Yes I am using(/loving) it. \n2) no it's not pytorch only. (link to documentation) \n3) it's like a online tensorboard which needs you to put something like \nwandb.log({\"loss\": loss}) in your code that's it :)\n
I just took a super quick look, but it seems like this might already be possible with Tensorboard, at least for people using Tensorflow, or am I missing something?\nSounds like a great addition anyways!
", "rawMarkdown": "I just took a super quick look, but it seems like this might already be possible with Tensorboard, at least for people using Tensorflow, or am I missing something?\nSounds like a great addition anyways!", "votes": null }, { "id": "640637", "postDate": "10/04/2019 06:23:18", "content": "Yes it is similar to Tensorboard but more capable. It is online so it does not worry about where you are running your experiments. AFAIK tensorboard should be running on same machine where logging is happening. so in tensorboard it is much difficult to plot the metrics from different runs (as in kaggle different run means different machine) in same plot as log needs to be at one place. I don' t know how will you use Tensorboard in kaggle kernels.
", "rawMarkdown": "Yes it is similar to Tensorboard but more capable. It is online so it does not worry about where you are running your experiments. AFAIK tensorboard should be running on same machine where logging is happening. so in tensorboard it is much difficult to plot the metrics from different runs (as in kaggle different run means different machine) in same plot as log needs to be at one place. I don' t know how will you use Tensorboard in kaggle kernels.", "votes": null }, { "id": "641025", "postDate": "10/04/2019 11:51:10", "content": "I used to use it, but it was indeed a bit tricky. You could create a link to open in a new browser tab that worked for as long as your kernel session was working.\nIt was pretty handy to get different metrics once the training was done, and to compare different models.\nHowever, you're right, this tool might be better inside Kaggle kernels. More options is always better anyways!
", "rawMarkdown": "I used to use it, but it was indeed a bit tricky. You could create a link to open in a new browser tab that worked for as long as your kernel session was working.\nIt was pretty handy to get different metrics once the training was done, and to compare different models.\nHowever, you're right, this tool might be better inside Kaggle kernels. More options is always better anyways!", "votes": null } ], "comments": [ { "id": 635760, "author_name": "mobassir", "author_url": "", "post_date": "09/28/2019 06:32:46", "content": "thank you a ton mate for this post,are you using it? is it only for pytorch users? @dhananjay3
", "votes": null, "replies": [ { "id": 635764, "author_name": "dhananjay3", "author_url": "", "post_date": "09/28/2019 06:49:48", "content": "1) Yes I am using(/loving) it. \n2) no it's not pytorch only. (link to documentation) \n3) it's like a online tensorboard which needs you to put something like \nwandb.log({\"loss\": loss}) in your code that's it :)\n
I just took a super quick look, but it seems like this might already be possible with Tensorboard, at least for people using Tensorflow, or am I missing something?\nSounds like a great addition anyways!
", "votes": null, "replies": [ { "id": 640637, "author_name": "dhananjay3", "author_url": "", "post_date": "10/04/2019 06:23:18", "content": "Yes it is similar to Tensorboard but more capable. It is online so it does not worry about where you are running your experiments. AFAIK tensorboard should be running on same machine where logging is happening. so in tensorboard it is much difficult to plot the metrics from different runs (as in kaggle different run means different machine) in same plot as log needs to be at one place. I don' t know how will you use Tensorboard in kaggle kernels.
", "votes": null, "replies": [] }, { "id": 641025, "author_name": "maxlenormand", "author_url": "", "post_date": "10/04/2019 11:51:10", "content": "I used to use it, but it was indeed a bit tricky. You could create a link to open in a new browser tab that worked for as long as your kernel session was working.\nIt was pretty handy to get different metrics once the training was done, and to compare different models.\nHowever, you're right, this tool might be better inside Kaggle kernels. More options is always better anyways!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "635733": "Do you ever wanted to see what is going on inside the code you just committed ? Even more due to recent GPU time limit where you don't want to waste your time in a interactive session. Try Weights and biases [(link)](https://www.wandb.com/). its super easy to setup even in kaggle kernel and you can see the log at their website. this way you can terminate your commit early if needed saving time plus it is free. you can keep track of your experiments like hyper parameters etc.\n\np.s. I am not any way connected to this company.", "635760": "thank you a ton mate for this post,are you using it? is it only for pytorch users? @dhananjay3", "635764": "1) Yes I am using(/loving) it. \n2) no it's not pytorch only. [(link to documentation)](https://docs.wandb.com/getting-started) \n3) it's like a online tensorboard which needs you to put something like \n`wandb.log({\"loss\": loss})` in your code that's it :)\n", "639760": "I just took a super quick look, but it seems like this might already be possible with Tensorboard, at least for people using Tensorflow, or am I missing something?\nSounds like a great addition anyways!", "640637": "Yes it is similar to Tensorboard but more capable. It is online so it does not worry about where you are running your experiments. AFAIK tensorboard should be running on same machine where logging is happening. so in tensorboard it is much difficult to plot the metrics from different runs (as in kaggle different run means different machine) in same plot as log needs to be at one place. I don' t know how will you use Tensorboard in kaggle kernels.", "641025": "I used to use it, but it was indeed a bit tricky. You could create a link to open in a new browser tab that worked for as long as your kernel session was working.\nIt was pretty handy to get different metrics once the training was done, and to compare different models.\nHowever, you're right, this tool might be better inside Kaggle kernels. More options is always better anyways!" }, "source": "meta" }