{ "id": 238130, "title": "[LB: 0.93+] Starter Notebook with Tensorflow |TPU πŸš€", "url": "/competitions/seti-breakthrough-listen/discussion/238130", "author_name": "Awsaf", "post_date": "2021-05-11T10:14:52.232000", "votes": 35, "comment_count": 3, "views": 0, "content": "

\"\"

\n

Train & Inference with Tensorflow TPU

\n

I've published a notebook using Tensorflow 2.0 & TPU.

\n

Notebooks:

\n\n

Datasets:

\n\n

NOTE:

\n\n
Not all of the β€œneedle” signals look like diagonal lines, and they may not be present for the entirety of all three β€œA” observations, but what they do have in common is that they are only present in some or all of the β€œA” observations (panels 1, 3, and 5 in the cadence snippets).\n
\n\n

Result

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
VersionCVLB
40.9300.91
60.9550.93
80.9490.92
\n

I hope you find this helpful and we may find some aliens at the end of this competition πŸ‘½

", "messages": [ { "id": 1301931, "postDate": "2021-05-11T10:14:52.233Z", "content": "

\"\"

\n

Train & Inference with Tensorflow TPU

\n

I've published a notebook using Tensorflow 2.0 & TPU.

\n

Notebooks:

\n\n

Datasets:

\n\n

NOTE:

\n\n
Not all of the β€œneedle” signals look like diagonal lines, and they may not be present for the entirety of all three β€œA” observations, but what they do have in common is that they are only present in some or all of the β€œA” observations (panels 1, 3, and 5 in the cadence snippets).\n
\n\n

Result

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
VersionCVLB
40.9300.91
60.9550.93
80.9490.92
\n

I hope you find this helpful and we may find some aliens at the end of this competition πŸ‘½

", "rawMarkdown": "![](https://assets.kpmg/is/image/kpmg/space-satelite-signal-receiving-antenas-at-night-banner:cq5dam.web.1400.350)\n## Train & Inference with Tensorflow TPU\nI've published a notebook using **Tensorflow 2.0** & **TPU**. \n\n### Notebooks:\n* **train [channel] :** [SETI-BL: TF Starter TPU πŸš€](https://www.kaggle.com/awsaf49/seti-bl-tf-starter-tpu) **LB:0.93+**\n* **train [spatial] :** [SETI-BL: Spatial Info [TF|TPU] πŸš€](https://www.kaggle.com/awsaf49/seti-bl-spatial-info-tf-tpu) **LB:0.95+**\n* **tfrecord dataset:** [SETI-BL: 256x256 tfrec Data](https://www.kaggle.com/awsaf49/seti-bl-256x256-tfrec-data)\n\n### Datasets:\n* [128x128](https://www.kaggle.com/awsaf49/setibl-128x128-tfrec-dataset)\n* [256x256](https://www.kaggle.com/awsaf49/setibl-256x256-tfrec-dataset)\n* [384x384](https://www.kaggle.com/awsaf49/setibl-384x384-tfrec-dataset)\n\n**NOTE:**\n* Original Signal had **6** channels. I only took **3** as it was mention here,\n```\nNot all of the β€œneedle” signals look like diagonal lines, and they may not be present for the entirety of all three β€œA” observations, but what they do have in common is that they are only present in some or all of the β€œA” observations (panels 1, 3, and 5 in the cadence snippets).\n```\n* Didn't do any **Signal Processing** so far. Left it for the model.\n* Used simple augmentations, some of them may hurt the model.\n* Resolution was kept **256x256** for the initial training\n* model: **EfficientNetB4**\n\n## Result\n| Version | CV | LB\n| --- | --- |\n| 4 | 0.930 | 0.91\n| 6 | 0.955 | 0.93\n| 8 | 0.949 | 0.92\n\n\nI hope you find this helpful and we may find some aliens at the end of this competition πŸ‘½\n\n", "votes": 35 }, { "id": 1302638, "postDate": "2021-05-11T16:33:16.063Z", "content": "

Wow that was fast!

", "rawMarkdown": "Wow that was fast!", "votes": 1 }, { "id": 1302579, "postDate": "2021-05-11T15:53:32.297Z", "content": "

Great work @awsaf49 !!! Awesome to see you get .90+ in the first day itself!!

\n

Kudos to you, … thanks for sharing and hope to see you edge closer to 1.00 πŸ‘πŸ––πŸ˜„β€οΈ

", "rawMarkdown": "Great work @awsaf49 !!! Awesome to see you get .90+ in the first day itself!!\n\nKudos to you, ... thanks for sharing and hope to see you edge closer to 1.00 πŸ‘πŸ––πŸ˜„β€οΈ", "votes": 1 }, { "id": 1302109, "postDate": "2021-05-11T11:50:38.400Z", "content": "

thanks for sharing a good pipeline!

", "rawMarkdown": "thanks for sharing a good pipeline!", "votes": 1 } ], "comments": [ { "id": 1302638, "author_name": "xhlulu", "author_url": "", "post_date": "2021-05-11T16:33:16.063000", "content": "

Wow that was fast!

", "votes": 1, "replies": [] }, { "id": 1302579, "author_name": "Kamal Das", "author_url": "", "post_date": "2021-05-11T15:53:32.297000", "content": "

Great work @awsaf49 !!! Awesome to see you get .90+ in the first day itself!!

\n

Kudos to you, … thanks for sharing and hope to see you edge closer to 1.00 πŸ‘πŸ––πŸ˜„β€οΈ

", "votes": 1, "replies": [] }, { "id": 1302109, "author_name": "Manh Lab", "author_url": "", "post_date": "2021-05-11T11:50:38.400000", "content": "

thanks for sharing a good pipeline!

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1301931": "![](https://assets.kpmg/is/image/kpmg/space-satelite-signal-receiving-antenas-at-night-banner:cq5dam.web.1400.350)\n## Train & Inference with Tensorflow TPU\nI've published a notebook using **Tensorflow 2.0** & **TPU**. \n\n### Notebooks:\n* **train [channel] :** [SETI-BL: TF Starter TPU πŸš€](https://www.kaggle.com/awsaf49/seti-bl-tf-starter-tpu) **LB:0.93+**\n* **train [spatial] :** [SETI-BL: Spatial Info [TF|TPU] πŸš€](https://www.kaggle.com/awsaf49/seti-bl-spatial-info-tf-tpu) **LB:0.95+**\n* **tfrecord dataset:** [SETI-BL: 256x256 tfrec Data](https://www.kaggle.com/awsaf49/seti-bl-256x256-tfrec-data)\n\n### Datasets:\n* [128x128](https://www.kaggle.com/awsaf49/setibl-128x128-tfrec-dataset)\n* [256x256](https://www.kaggle.com/awsaf49/setibl-256x256-tfrec-dataset)\n* [384x384](https://www.kaggle.com/awsaf49/setibl-384x384-tfrec-dataset)\n\n**NOTE:**\n* Original Signal had **6** channels. I only took **3** as it was mention here,\n```\nNot all of the β€œneedle” signals look like diagonal lines, and they may not be present for the entirety of all three β€œA” observations, but what they do have in common is that they are only present in some or all of the β€œA” observations (panels 1, 3, and 5 in the cadence snippets).\n```\n* Didn't do any **Signal Processing** so far. Left it for the model.\n* Used simple augmentations, some of them may hurt the model.\n* Resolution was kept **256x256** for the initial training\n* model: **EfficientNetB4**\n\n## Result\n| Version | CV | LB\n| --- | --- |\n| 4 | 0.930 | 0.91\n| 6 | 0.955 | 0.93\n| 8 | 0.949 | 0.92\n\n\nI hope you find this helpful and we may find some aliens at the end of this competition πŸ‘½\n\n", "1302638": "Wow that was fast!", "1302579": "Great work @awsaf49 !!! Awesome to see you get .90+ in the first day itself!!\n\nKudos to you, ... thanks for sharing and hope to see you edge closer to 1.00 πŸ‘πŸ––πŸ˜„β€οΈ", "1302109": "thanks for sharing a good pipeline!" } }