Experimental design with small dataset
$begingroup$
I am working with small medical images dataset, my goal is to classify images into categories. The dataset classes have rather big intra class variation, as on some of the images broken bones, screws, plates etc. are present. That's why I am using automatic split with label stratifying, rather than one common split, during my experiments. However, as biggest class has around 2k of images and smallest around 60, it is probably not small enough for OLOCV to make sense.
I would like to create experiments, that make sense on one hand and have value on the other one. At the moment, I am splitting dataset 70:15:15, saving the results and training network. I have thought about creating 30 splits, trying to evaluate them on some fast network for few epochs (let's say 20), selecting 10 that perform the best and running them through all networks I am using for transfer learning. Then, finding the best performing network and try to optimize it with SGD ( I am using Adam atm) as some literature suggest that SGD is generalizing better than Adam.
I would then state best performance and average performance amongst the networks and best performance of neural network with SGD optimized on the best train/val/test split.
Does this make sense or do I have any big logical error in my way of thinking? Is there any other methodology you may please suggest to me, or any brief literature that may suggest such methodology?
Thank you.
//Edit: I would run the 10 splits through all the networks for 300 or 400 epochs.
deep-learning experiments
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
add a comment |
$begingroup$
I am working with small medical images dataset, my goal is to classify images into categories. The dataset classes have rather big intra class variation, as on some of the images broken bones, screws, plates etc. are present. That's why I am using automatic split with label stratifying, rather than one common split, during my experiments. However, as biggest class has around 2k of images and smallest around 60, it is probably not small enough for OLOCV to make sense.
I would like to create experiments, that make sense on one hand and have value on the other one. At the moment, I am splitting dataset 70:15:15, saving the results and training network. I have thought about creating 30 splits, trying to evaluate them on some fast network for few epochs (let's say 20), selecting 10 that perform the best and running them through all networks I am using for transfer learning. Then, finding the best performing network and try to optimize it with SGD ( I am using Adam atm) as some literature suggest that SGD is generalizing better than Adam.
I would then state best performance and average performance amongst the networks and best performance of neural network with SGD optimized on the best train/val/test split.
Does this make sense or do I have any big logical error in my way of thinking? Is there any other methodology you may please suggest to me, or any brief literature that may suggest such methodology?
Thank you.
//Edit: I would run the 10 splits through all the networks for 300 or 400 epochs.
deep-learning experiments
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
add a comment |
$begingroup$
I am working with small medical images dataset, my goal is to classify images into categories. The dataset classes have rather big intra class variation, as on some of the images broken bones, screws, plates etc. are present. That's why I am using automatic split with label stratifying, rather than one common split, during my experiments. However, as biggest class has around 2k of images and smallest around 60, it is probably not small enough for OLOCV to make sense.
I would like to create experiments, that make sense on one hand and have value on the other one. At the moment, I am splitting dataset 70:15:15, saving the results and training network. I have thought about creating 30 splits, trying to evaluate them on some fast network for few epochs (let's say 20), selecting 10 that perform the best and running them through all networks I am using for transfer learning. Then, finding the best performing network and try to optimize it with SGD ( I am using Adam atm) as some literature suggest that SGD is generalizing better than Adam.
I would then state best performance and average performance amongst the networks and best performance of neural network with SGD optimized on the best train/val/test split.
Does this make sense or do I have any big logical error in my way of thinking? Is there any other methodology you may please suggest to me, or any brief literature that may suggest such methodology?
Thank you.
//Edit: I would run the 10 splits through all the networks for 300 or 400 epochs.
deep-learning experiments
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
I am working with small medical images dataset, my goal is to classify images into categories. The dataset classes have rather big intra class variation, as on some of the images broken bones, screws, plates etc. are present. That's why I am using automatic split with label stratifying, rather than one common split, during my experiments. However, as biggest class has around 2k of images and smallest around 60, it is probably not small enough for OLOCV to make sense.
I would like to create experiments, that make sense on one hand and have value on the other one. At the moment, I am splitting dataset 70:15:15, saving the results and training network. I have thought about creating 30 splits, trying to evaluate them on some fast network for few epochs (let's say 20), selecting 10 that perform the best and running them through all networks I am using for transfer learning. Then, finding the best performing network and try to optimize it with SGD ( I am using Adam atm) as some literature suggest that SGD is generalizing better than Adam.
I would then state best performance and average performance amongst the networks and best performance of neural network with SGD optimized on the best train/val/test split.
Does this make sense or do I have any big logical error in my way of thinking? Is there any other methodology you may please suggest to me, or any brief literature that may suggest such methodology?
Thank you.
//Edit: I would run the 10 splits through all the networks for 300 or 400 epochs.
deep-learning experiments
deep-learning experiments
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
asked 13 hours ago
sob3kxsob3kx
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
asked 13 hours ago
sob3kxsob3kx
1
asked 13 hours ago
sob3kxsob3kx
1
1
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
New contributor
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
sob3kx is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
add a comment |
add a comment |
0
active
oldest
votes
Your Answer
StackExchange.ifUsing("editor", function () {
return StackExchange.using("mathjaxEditing", function () {
StackExchange.MarkdownEditor.creationCallbacks.add(function (editor, postfix) {
StackExchange.mathjaxEditing.prepareWmdForMathJax(editor, postfix, [["$", "$"], ["\\(","\\)"]]);
});
});
}, "mathjax-editing");
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "557"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});
function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: false,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: null,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});
}
});
sob3kx is a new contributor. Be nice, and check out our Code of Conduct.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fdatascience.stackexchange.com%2fquestions%2f46576%2fexperimental-design-with-small-dataset%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
0
active
oldest
votes
0
active
oldest
votes
active
oldest
votes
active
oldest
votes
sob3kx is a new contributor. Be nice, and check out our Code of Conduct.
sob3kx is a new contributor. Be nice, and check out our Code of Conduct.
sob3kx is a new contributor. Be nice, and check out our Code of Conduct.
sob3kx is a new contributor. Be nice, and check out our Code of Conduct.
Thanks for contributing an answer to Data Science Stack Exchange!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
Use MathJax to format equations. MathJax reference.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fdatascience.stackexchange.com%2fquestions%2f46576%2fexperimental-design-with-small-dataset%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
