![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2599569%2Fc8de93dc2e4eea126d3654144aa82113%2Flabellll.png?generation=1607673525163097&alt=media\")
As we all have noticed that there is a significant bias in the class label distribution, in our training dataset :
\n
Hence, sampling our dataset for validation becomes important(there isn't a validation set provided), and a random sample won't do effectively because of the aforementioned bias.
\nEnter : Stratified Sampling
\nThis method, which is a form of random sampling, consists of dividing the entire population being studied into different subgroups(here, according to the classes of diseases), so that an individual can belong to only one class(the singular). Once the groups have been defined, in order to create a sample, we select individuals by applying a sampling method to each of the groups separately.
\nThis notebook samples the data using stratified sampling, followed by the creation of TRAINING, VALIDATION & TESTING TFRECORDS. :
\nhttps://www.kaggle.com/fireheart7/cassava-create-tfrecords
\nA 256 X 256 record prepared(without any image preprocessing) is available at :
\nhttps://www.kaggle.com/fireheart7/cassava-tfrecords-version-neptune
", "messages": [ { "id": 1108981, "postDate": "2020-12-11T08:07:50.117Z", "content": "As we all have noticed that there is a significant bias in the class label distribution, in our training dataset :
\n
Hence, sampling our dataset for validation becomes important(there isn't a validation set provided), and a random sample won't do effectively because of the aforementioned bias.
\nEnter : Stratified Sampling
\nThis method, which is a form of random sampling, consists of dividing the entire population being studied into different subgroups(here, according to the classes of diseases), so that an individual can belong to only one class(the singular). Once the groups have been defined, in order to create a sample, we select individuals by applying a sampling method to each of the groups separately.
\nThis notebook samples the data using stratified sampling, followed by the creation of TRAINING, VALIDATION & TESTING TFRECORDS. :
\nhttps://www.kaggle.com/fireheart7/cassava-create-tfrecords
\nA 256 X 256 record prepared(without any image preprocessing) is available at :
\nhttps://www.kaggle.com/fireheart7/cassava-tfrecords-version-neptune
", "rawMarkdown": "As we all have noticed that there is a significant bias in the class label distribution, in our training dataset :\n\n\n\nHence, sampling our dataset for validation becomes important(*there isn't a validation set provided*), and a random sample won't do effectively because of the aforementioned bias.\n\n**Enter : Stratified Sampling**\n\nThis method, which is a form of random sampling, consists of dividing the entire population being studied into different subgroups(**here, according to the classes of diseases**), so that an individual can belong to only one class(the singular). Once the groups have been defined, in order to create a sample, we select individuals by applying a sampling method to each of the groups separately.\n\n**This notebook samples the data using stratified sampling, followed by the creation of TRAINING, VALIDATION & TESTING TFRECORDS**. : \n\n**https://www.kaggle.com/fireheart7/cassava-create-tfrecords**\n\nA 256 X 256 record prepared(without any image preprocessing) is available at : \n\n**https://www.kaggle.com/fireheart7/cassava-tfrecords-version-neptune** ", "votes": 6 } ], "comments": [], "raw_markdown_by_id": { "1108981": "As we all have noticed that there is a significant bias in the class label distribution, in our training dataset :\n\n\n\nHence, sampling our dataset for validation becomes important(*there isn't a validation set provided*), and a random sample won't do effectively because of the aforementioned bias.\n\n**Enter : Stratified Sampling**\n\nThis method, which is a form of random sampling, consists of dividing the entire population being studied into different subgroups(**here, according to the classes of diseases**), so that an individual can belong to only one class(the singular). Once the groups have been defined, in order to create a sample, we select individuals by applying a sampling method to each of the groups separately.\n\n**This notebook samples the data using stratified sampling, followed by the creation of TRAINING, VALIDATION & TESTING TFRECORDS**. : \n\n**https://www.kaggle.com/fireheart7/cassava-create-tfrecords**\n\nA 256 X 256 record prepared(without any image preprocessing) is available at : \n\n**https://www.kaggle.com/fireheart7/cassava-tfrecords-version-neptune** " } }