{ "id": 250278, "title": "G2Net starter code [Single Fold LB: 0.864]", "url": "/competitions/g2net-gravitational-wave-detection/discussion/250278", "author_name": "Y.Nakama", "post_date": "2021-07-02T00:53:29.735000", "votes": 107, "comment_count": 12, "views": 0, "content": "

I prepared starter code for this competition.

\n

[efficientnet_b0 melspectrogram approach]

\n\n

[efficientnet_b0 Q-transform approach]

\n\n

[efficientnet_b7 Q-transform approach]

\n\n

[efficientnet_b7 Q-transform approach version2]

\n\n

Hope this helps, happy kaggling!

", "messages": [ { "id": 1372727, "postDate": "2021-07-02T00:53:29.737Z", "content": "

I prepared starter code for this competition.

\n

[efficientnet_b0 melspectrogram approach]

\n\n

[efficientnet_b0 Q-transform approach]

\n\n

[efficientnet_b7 Q-transform approach]

\n\n

[efficientnet_b7 Q-transform approach version2]

\n\n

Hope this helps, happy kaggling!

", "rawMarkdown": "I prepared starter code for this competition.\n\n[efficientnet_b0 melspectrogram approach]\n- [train images dataset](https://www.kaggle.com/yasufuminakama/g2net-n-mels-128-train-images) & [generation code](https://www.kaggle.com/yasufuminakama/g2net-spectrogram-generation-train)\n- [test images dataset](https://www.kaggle.com/yasufuminakama/g2net-n-mels-128-test-images) & [generation code](https://www.kaggle.com/yasufuminakama/g2net-spectrogram-generation-test)\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-training?scriptVersionId=67215906)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-inference?scriptVersionId=67228556)\n- The result is CV: 0.819, LB: 0.830.\n- There may be other better preprocessing...\n\n[efficientnet_b0 Q-transform approach]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-training?scriptVersionId=67393876)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-inference?scriptVersionId=67400100)\n- The result is Single Fold: 0.851, LB: 0.856.\n\n[efficientnet_b7 Q-transform approach]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training?scriptVersionId=67408619)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-inference?scriptVersionId=67420856)\n- The result is Single Fold: 0.857, LB: 0.860.\n\n[efficientnet_b7 Q-transform approach version2]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training?scriptVersionId=69947710)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-inference?scriptVersionId=70010612)\n- The result is Single Fold: 0.861, LB: 0.864.\n\nHope this helps, happy kaggling!", "votes": 103 }, { "id": 1373060, "postDate": "2021-07-02T07:54:00.967Z", "content": "

@yasufuminakama cool job as always!
\nI have a question, why are you using sr == 4096 when converting to melspec?

\n
melspec = librosa.feature.melspectrogram(\n    waves[j] / max(waves[j]), sr=4096, n_mels=128, fmin=20, fmax=2048\n)\n
\n

I thought the sample rate of the signals is equaled 2048

\n
\n

Each data sample (npy file) contains 3 time series (1 for each detector) and each spans 2 sec and is sampled at 2,048 Hz.

\n
", "rawMarkdown": "@yasufuminakama cool job as always!\nI have a question, why are you using sr == 4096 when converting to melspec?\n\n```\nmelspec = librosa.feature.melspectrogram(\n waves[j] / max(waves[j]), sr=4096, n_mels=128, fmin=20, fmax=2048\n)\n```\n\nI thought the sample rate of the signals is equaled 2048\n\n> Each data sample (npy file) contains 3 time series (1 for each detector) and each spans 2 sec and is sampled at 2,048 Hz.", "votes": 5, "replies": [ { "id": 1373066, "postDate": "2021-07-02T08:02:39.870Z", "content": "

Thanks!
\nI just missed it… and now I think q-transformed images such as https://www.kaggle.com/alexnitz/pycbc-making-images will work better than melspectrogram, I will update the code & results.

", "rawMarkdown": "Thanks!\nI just missed it... and now I think q-transformed images such as https://www.kaggle.com/alexnitz/pycbc-making-images will work better than melspectrogram, I will update the code & results.", "votes": 6 }, { "id": 1373597, "postDate": "2021-07-02T16:06:14.343Z", "content": "

I created the dataset by changing sr=2048 in the @Y.Nakama's original notebook.

\n\n


\ndone!

", "rawMarkdown": "I created the dataset by changing `sr=2048` in the @Y.Nakama's original notebook.\n\n- [train dataset link for sr=2048](https://www.kaggle.com/piantic/g2net-n-mels-128-sr-2048-train-images) | [train dataset generation code for sr=2048](https://www.kaggle.com/piantic/g2net-spectrogram-generation-train-sr-2048?scriptVersionId=67278253)\n\n\n- [test dataset link for sr=2048](https://www.kaggle.com/piantic/g2net-n-mels-128-sr-2048-test-images) | [test dataset generation code for sr=2048](https://www.kaggle.com/piantic/g2net-spectrogram-generation-test-sr-2048)\n\n\n\n\n~~p.s. The generation code has been running for train. I will update it later.~~\ndone!", "votes": 4 }, { "id": 1373815, "postDate": "2021-07-02T18:16:11.803Z", "content": "

The dataset I shared above is sr=2048. It seems that it would be better to also change fmax to 1024.

\n\n

p.s. The above dataset will be updated again.

", "rawMarkdown": "The dataset I shared above is sr=2048. It seems that it would be better to also change `fmax` to `1024`.\n\n- `sr=2048`\n![](https://imgur.com/rIcOH4S.jpg)\n\n- `sr=2048` & `fmax=1024`\n![](https://imgur.com/xBd9mkF.jpg)\n\np.s. The above dataset will be updated again.\n\n", "votes": 2 }, { "id": 1374686, "postDate": "2021-07-03T14:09:32.617Z", "rawMarkdown": "", "votes": 2, "isDeleted": true } ] }, { "id": 1372757, "postDate": "2021-07-02T01:58:49.563Z", "content": "

Really helpful resource, Thank you for sharing

", "rawMarkdown": "Really helpful resource, Thank you for sharing", "votes": 1 }, { "id": 1373139, "postDate": "2021-07-02T08:42:45.337Z", "content": "

Your starter code is very helpful for beginners like me😊
\nI'll try to make my own notebook while referring to it.

\n

Thanks🪐

", "rawMarkdown": "Your starter code is very helpful for beginners like me😊\nI'll try to make my own notebook while referring to it.\n\nThanks🪐", "votes": 2 }, { "id": 1563253, "postDate": "2021-10-28T06:51:48.647Z", "content": "

Hey,

\n

Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.

\n

Thank you all,

\n

Regards,
\nChris

", "rawMarkdown": "Hey,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris" }, { "id": 1374820, "postDate": "2021-07-03T16:40:00.967Z", "content": "

I am new and I am stuck at the data reading part: train = pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv'). I got error: FileNotFoundError: [Errno 2] No such file or directory: '../input/g2net-gravitational-wave-detection/training_labels.csv'. Can someone help me on how do I access the data? I am using the Kaggle kernel.

", "rawMarkdown": "I am new and I am stuck at the data reading part: train = pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv'). I got error: FileNotFoundError: [Errno 2] No such file or directory: '../input/g2net-gravitational-wave-detection/training_labels.csv'. Can someone help me on how do I access the data? I am using the Kaggle kernel.", "replies": [ { "id": 1374826, "postDate": "2021-07-03T16:46:46.590Z", "content": "

Make sure you have added the competition's data like this.

\n
import os\nprint(os.listdir('../input/'))\n
\n

You can add the competition's data from + Add data for your kaggle kernel.

", "rawMarkdown": "Make sure you have added the competition's data like this.\n```\nimport os\nprint(os.listdir('../input/'))\n```\nYou can add the competition's data from `+ Add data` for your kaggle kernel.", "votes": 1 }, { "id": 1374843, "postDate": "2021-07-03T16:56:42.120Z", "content": "

Thanks Y.Nakama. I added the competition data and now it works. Thanks for you help.

", "rawMarkdown": "Thanks Y.Nakama. I added the competition data and now it works. Thanks for you help.", "votes": 1 } ] }, { "id": 1373448, "postDate": "2021-07-02T13:50:29.213Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1373060, "author_name": "Yaroslav Isaienkov", "author_url": "", "post_date": "2021-07-02T07:54:00.967000", "content": "

@yasufuminakama cool job as always!
\nI have a question, why are you using sr == 4096 when converting to melspec?

\n
melspec = librosa.feature.melspectrogram(\n    waves[j] / max(waves[j]), sr=4096, n_mels=128, fmin=20, fmax=2048\n)\n
\n

I thought the sample rate of the signals is equaled 2048

\n
\n

Each data sample (npy file) contains 3 time series (1 for each detector) and each spans 2 sec and is sampled at 2,048 Hz.

\n
", "votes": 5, "replies": [ { "id": 1373066, "author_name": "Y.Nakama", "author_url": "", "post_date": "2021-07-02T08:02:39.870000", "content": "

Thanks!
\nI just missed it… and now I think q-transformed images such as https://www.kaggle.com/alexnitz/pycbc-making-images will work better than melspectrogram, I will update the code & results.

", "votes": 6, "replies": [] }, { "id": 1373597, "author_name": "Heroseo", "author_url": "", "post_date": "2021-07-02T16:06:14.343000", "content": "

I created the dataset by changing sr=2048 in the @Y.Nakama's original notebook.

\n\n


\ndone!

", "votes": 4, "replies": [] }, { "id": 1373815, "author_name": "Heroseo", "author_url": "", "post_date": "2021-07-02T18:16:11.803000", "content": "

The dataset I shared above is sr=2048. It seems that it would be better to also change fmax to 1024.

\n\n

p.s. The above dataset will be updated again.

", "votes": 2, "replies": [] }, { "id": 1374686, "author_name": "", "author_url": "", "post_date": "2021-07-03T14:09:32.617000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 1372757, "author_name": "Samarth Gupta", "author_url": "", "post_date": "2021-07-02T01:58:49.563000", "content": "

Really helpful resource, Thank you for sharing

", "votes": 1, "replies": [] }, { "id": 1373139, "author_name": "shokupan", "author_url": "", "post_date": "2021-07-02T08:42:45.337000", "content": "

Your starter code is very helpful for beginners like me😊
\nI'll try to make my own notebook while referring to it.

\n

Thanks🪐

", "votes": 2, "replies": [] }, { "id": 1563253, "author_name": "ChristopherZerafa", "author_url": "", "post_date": "2021-10-28T06:51:48.647000", "content": "

Hey,

\n

Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.

\n

Thank you all,

\n

Regards,
\nChris

", "votes": 0, "replies": [] }, { "id": 1374820, "author_name": "Dongthias", "author_url": "", "post_date": "2021-07-03T16:40:00.967000", "content": "

I am new and I am stuck at the data reading part: train = pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv'). I got error: FileNotFoundError: [Errno 2] No such file or directory: '../input/g2net-gravitational-wave-detection/training_labels.csv'. Can someone help me on how do I access the data? I am using the Kaggle kernel.

", "votes": 0, "replies": [ { "id": 1374826, "author_name": "Y.Nakama", "author_url": "", "post_date": "2021-07-03T16:46:46.590000", "content": "

Make sure you have added the competition's data like this.

\n
import os\nprint(os.listdir('../input/'))\n
\n

You can add the competition's data from + Add data for your kaggle kernel.

", "votes": 1, "replies": [] }, { "id": 1374843, "author_name": "Dongthias", "author_url": "", "post_date": "2021-07-03T16:56:42.120000", "content": "

Thanks Y.Nakama. I added the competition data and now it works. Thanks for you help.

", "votes": 1, "replies": [] } ] }, { "id": 1373448, "author_name": "", "author_url": "", "post_date": "2021-07-02T13:50:29.213000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1372727": "I prepared starter code for this competition.\n\n[efficientnet_b0 melspectrogram approach]\n- [train images dataset](https://www.kaggle.com/yasufuminakama/g2net-n-mels-128-train-images) & [generation code](https://www.kaggle.com/yasufuminakama/g2net-spectrogram-generation-train)\n- [test images dataset](https://www.kaggle.com/yasufuminakama/g2net-n-mels-128-test-images) & [generation code](https://www.kaggle.com/yasufuminakama/g2net-spectrogram-generation-test)\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-training?scriptVersionId=67215906)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-inference?scriptVersionId=67228556)\n- The result is CV: 0.819, LB: 0.830.\n- There may be other better preprocessing...\n\n[efficientnet_b0 Q-transform approach]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-training?scriptVersionId=67393876)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b0-baseline-inference?scriptVersionId=67400100)\n- The result is Single Fold: 0.851, LB: 0.856.\n\n[efficientnet_b7 Q-transform approach]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training?scriptVersionId=67408619)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-inference?scriptVersionId=67420856)\n- The result is Single Fold: 0.857, LB: 0.860.\n\n[efficientnet_b7 Q-transform approach version2]\n- [training](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training?scriptVersionId=69947710)\n- [inference](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-inference?scriptVersionId=70010612)\n- The result is Single Fold: 0.861, LB: 0.864.\n\nHope this helps, happy kaggling!", "1373060": "@yasufuminakama cool job as always!\nI have a question, why are you using sr == 4096 when converting to melspec?\n\n```\nmelspec = librosa.feature.melspectrogram(\n waves[j] / max(waves[j]), sr=4096, n_mels=128, fmin=20, fmax=2048\n)\n```\n\nI thought the sample rate of the signals is equaled 2048\n\n> Each data sample (npy file) contains 3 time series (1 for each detector) and each spans 2 sec and is sampled at 2,048 Hz.", "1372757": "Really helpful resource, Thank you for sharing", "1373139": "Your starter code is very helpful for beginners like me😊\nI'll try to make my own notebook while referring to it.\n\nThanks🪐", "1563253": "Hey,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris", "1374820": "I am new and I am stuck at the data reading part: train = pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv'). I got error: FileNotFoundError: [Errno 2] No such file or directory: '../input/g2net-gravitational-wave-detection/training_labels.csv'. Can someone help me on how do I access the data? I am using the Kaggle kernel.", "1373448": "" } }