{ "id": 246182, "title": "[Doubt]Question About Making predictions on Hidden Test Set", "url": "/competitions/siim-covid19-detection/discussion/246182", "author_name": "Athar Sayed", "post_date": "2021-06-14T11:22:16.849000", "votes": 5, "comment_count": 3, "views": 0, "content": "

I am using @ammarnassanalhajali this notebook to make Test Inference.
\nAs Far as I understand this notebook takes the preprocessed Test Image Data Set of size like 512 or 384 or 640 and make inference but , the competition involves making predictions on Hidden Test set which is not available publicly.
\nSo I am thinking that if we use this notebook directly there will be problem while making inference on test set as some of this preprocessed files may not be present on the public Lb .
\nPlease make me correct if I am mistaken here

", "messages": [ { "id": 1348898, "postDate": "2021-06-14T11:22:16.850Z", "content": "

I am using @ammarnassanalhajali this notebook to make Test Inference.
\nAs Far as I understand this notebook takes the preprocessed Test Image Data Set of size like 512 or 384 or 640 and make inference but , the competition involves making predictions on Hidden Test set which is not available publicly.
\nSo I am thinking that if we use this notebook directly there will be problem while making inference on test set as some of this preprocessed files may not be present on the public Lb .
\nPlease make me correct if I am mistaken here

", "rawMarkdown": "I am using @ammarnassanalhajali [this ](https://www.kaggle.com/ammarnassanalhajali/siim-covid-19-detectron2-inferance)notebook to make Test Inference.\nAs Far as I understand this notebook takes the preprocessed Test Image Data Set of size like 512 or 384 or 640 and make inference but , the competition involves making predictions on Hidden Test set which is not available publicly.\nSo I am thinking that if we use this notebook directly there will be problem while making inference on test set as some of this preprocessed files may not be present on the public Lb .\nPlease make me correct if I am mistaken here ", "votes": 5 }, { "id": 1349187, "postDate": "2021-06-14T15:16:06.030Z", "content": "

Yes. Your submission notebook should be able to read all image files from test directory and preprocess them as you did during training and then make inference on them. As you mentioned rightly, new images are added to the test directory during LB score evaluation. So, using predefined set of images only will not work. The preprocessed test images available however can be of help to make sure your model is working on them correctly.

", "rawMarkdown": "Yes. Your submission notebook should be able to read all image files from test directory and preprocess them as you did during training and then make inference on them. As you mentioned rightly, new images are added to the test directory during LB score evaluation. So, using predefined set of images only will not work. The preprocessed test images available however can be of help to make sure your model is working on them correctly.", "votes": 1 }, { "id": 1385127, "postDate": "2021-07-12T13:28:32.730Z", "content": "

so we should make inference from test folder or sample submission.csv?

", "rawMarkdown": "so we should make inference from test folder or sample submission.csv? " }, { "id": 1351688, "postDate": "2021-06-16T14:22:22.350Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1349187, "author_name": "Raja Suman C", "author_url": "", "post_date": "2021-06-14T15:16:06.030000", "content": "

Yes. Your submission notebook should be able to read all image files from test directory and preprocess them as you did during training and then make inference on them. As you mentioned rightly, new images are added to the test directory during LB score evaluation. So, using predefined set of images only will not work. The preprocessed test images available however can be of help to make sure your model is working on them correctly.

", "votes": 1, "replies": [] }, { "id": 1385127, "author_name": "Haw Keat", "author_url": "", "post_date": "2021-07-12T13:28:32.730000", "content": "

so we should make inference from test folder or sample submission.csv?

", "votes": 0, "replies": [] }, { "id": 1351688, "author_name": "", "author_url": "", "post_date": "2021-06-16T14:22:22.350000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1348898": "I am using @ammarnassanalhajali [this ](https://www.kaggle.com/ammarnassanalhajali/siim-covid-19-detectron2-inferance)notebook to make Test Inference.\nAs Far as I understand this notebook takes the preprocessed Test Image Data Set of size like 512 or 384 or 640 and make inference but , the competition involves making predictions on Hidden Test set which is not available publicly.\nSo I am thinking that if we use this notebook directly there will be problem while making inference on test set as some of this preprocessed files may not be present on the public Lb .\nPlease make me correct if I am mistaken here ", "1349187": "Yes. Your submission notebook should be able to read all image files from test directory and preprocess them as you did during training and then make inference on them. As you mentioned rightly, new images are added to the test directory during LB score evaluation. So, using predefined set of images only will not work. The preprocessed test images available however can be of help to make sure your model is working on them correctly.", "1385127": "so we should make inference from test folder or sample submission.csv? ", "1351688": "" } }