![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3721289%2F1440882a0026863f699a5993d3f985b2%2F.png?generation=1570457980325359&alt=media\")
We'd like to remind you what will happen in the second stage of this competition:
\n\nWhy second stage? Why so much trouble?
\n\nThe spirit of having a second stage is to prevent hand labeling and leaderboard probing of the test data. In order to achieve this, we ask that you upload your source code, including the correct parameters that you used for generating your submission files. This is for you to prove that you have written automated code to create your final submission(s). These \"models\" that you submit may be examined by Kaggle and the competition host to determine your eligibility to win the competition and claim prizes.
\n\nIf I don't have a chance to win, should I upload my model?
\n\nYes. You never know where your team will place on private leaderboard, so it’s worth uploading.
\n\nWill I still be on the leaderboard if I don't submit in the second stage?
\n\nYou will not. You will need to make a submission in the correct format in the second stage to remain on the leaderboard.
\n\nWill I still be on the leaderboard if I don't upload a model but do submit in the second stage?
\n\nYes. However, if you do not upload a model and finish in a prize position, your team will be removed from the competition standings entirely!
\n\nHow do I upload my model?
\n\nTo ensure that you did write code to produce your results, you are requested to upload your model. To do this, you can go to \"More->Team->Your Model\" and upload an archive of your code. It is necessary to zip everything as a single file. Please note that this upload link becomes unavailable after the deadline of stage 1, so you will need to upload it before the end of stage 1.
\n\nWhat should I upload?
\n\nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those. However, you should submit the code used to generate those files. You can typically select two submissions for final scoring, so don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each.
\n\nWhat happens to my pre-trained model?
\n\nYou only need to include a README file to indicate where you can download the pre-trained files from. For example, if you used vgg16 from keras, you don’t need to upload the weights file, you only need to indicate where you got it from: https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5
\n\nWhat if my submission is too big?
\n\nOur uploader will handle reasonably large files. If you still think your model will be too large, you can instead upload a checksum of your archive file (such as an md5 or sha hash). Note that if you do win, you will still have to upload it. A common reason for folder being too big might be that you included too many of your non-code files. If you upload a checksum, finish in a prize spot, and are unable to subsequently provide an archive that matches the checksum, you will be removed from the competition.
\n\nWhat if I want my code to stay private if I don’t win
\n\nYou may upload an encrypted archive and provide the decryption key in the event you win and wish to claim a prize. Alternatively, you can use a checksum, as described above.
\n\nWhat happens if I want to change something in my code in the second stage?
\n\nWe expect you may need to make some “non scientific” alterations, such as changes to path names, in order to create your submissions for the second stage. You are allowed to re-train your model (including the stage one data), but your code should not change. You should not be doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long as it is fully automated.
\n\nCan I upload multiple times?
\n\nYou may upload as many model files as you wish. Kaggle only keeps the latest upload. Make sure your last upload is the right one!
\n\nWill the number of participants change in the second stage? Will I get a medal? Will I get points?
\n\nYes, the number of participants will be smaller since some people won't submit in the second stage. For the purposes of medals and points calculations, we will use the number of participants in the first stage to calculate points and medals.
", "messages": [ { "id": 180543, "postDate": "2017-05-05T20:26:33.280Z", "content": "We'd like to remind you what will happen in the second stage of this competition:
\n\nWhy second stage? Why so much trouble?
\n\nThe spirit of having a second stage is to prevent hand labeling and leaderboard probing of the test data. In order to achieve this, we ask that you upload your source code, including the correct parameters that you used for generating your submission files. This is for you to prove that you have written automated code to create your final submission(s). These \"models\" that you submit may be examined by Kaggle and the competition host to determine your eligibility to win the competition and claim prizes.
\n\nIf I don't have a chance to win, should I upload my model?
\n\nYes. You never know where your team will place on private leaderboard, so it’s worth uploading.
\n\nWill I still be on the leaderboard if I don't submit in the second stage?
\n\nYou will not. You will need to make a submission in the correct format in the second stage to remain on the leaderboard.
\n\nWill I still be on the leaderboard if I don't upload a model but do submit in the second stage?
\n\nYes. However, if you do not upload a model and finish in a prize position, your team will be removed from the competition standings entirely!
\n\nHow do I upload my model?
\n\nTo ensure that you did write code to produce your results, you are requested to upload your model. To do this, you can go to \"More->Team->Your Model\" and upload an archive of your code. It is necessary to zip everything as a single file. Please note that this upload link becomes unavailable after the deadline of stage 1, so you will need to upload it before the end of stage 1.
\n\nWhat should I upload?
\n\nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those. However, you should submit the code used to generate those files. You can typically select two submissions for final scoring, so don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each.
\n\nWhat happens to my pre-trained model?
\n\nYou only need to include a README file to indicate where you can download the pre-trained files from. For example, if you used vgg16 from keras, you don’t need to upload the weights file, you only need to indicate where you got it from: https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5
\n\nWhat if my submission is too big?
\n\nOur uploader will handle reasonably large files. If you still think your model will be too large, you can instead upload a checksum of your archive file (such as an md5 or sha hash). Note that if you do win, you will still have to upload it. A common reason for folder being too big might be that you included too many of your non-code files. If you upload a checksum, finish in a prize spot, and are unable to subsequently provide an archive that matches the checksum, you will be removed from the competition.
\n\nWhat if I want my code to stay private if I don’t win
\n\nYou may upload an encrypted archive and provide the decryption key in the event you win and wish to claim a prize. Alternatively, you can use a checksum, as described above.
\n\nWhat happens if I want to change something in my code in the second stage?
\n\nWe expect you may need to make some “non scientific” alterations, such as changes to path names, in order to create your submissions for the second stage. You are allowed to re-train your model (including the stage one data), but your code should not change. You should not be doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long as it is fully automated.
\n\nCan I upload multiple times?
\n\nYou may upload as many model files as you wish. Kaggle only keeps the latest upload. Make sure your last upload is the right one!
\n\nWill the number of participants change in the second stage? Will I get a medal? Will I get points?
\n\nYes, the number of participants will be smaller since some people won't submit in the second stage. For the purposes of medals and points calculations, we will use the number of participants in the first stage to calculate points and medals.
", "rawMarkdown": "We'd like to remind you what will happen in the second stage of this competition:\n \n \n**Why second stage? Why so much trouble?**\n \nThe spirit of having a second stage is to prevent hand labeling and leaderboard probing of the test data. In order to achieve this, we ask that you upload your source code, including the correct parameters that you used for generating your submission files. This is for you to prove that you have written automated code to create your final submission(s). These \"models\" that you submit may be examined by Kaggle and the competition host to determine your eligibility to win the competition and claim prizes.\n \n**If I don't have a chance to win, should I upload my model?**\n \nYes. You never know where your team will place on private leaderboard, so it’s worth uploading.\n \n**Will I still be on the leaderboard if I don't submit in the second stage?**\n \nYou will not. You will need to make a submission in the correct format in the second stage to remain on the leaderboard. \n \n**Will I still be on the leaderboard if I don't upload a model but do submit in the second stage?**\n \nYes. However, if you do not upload a model and finish in a prize position, your team will be removed from the competition standings entirely!\n \n**How do I upload my model?** \n \nTo ensure that you did write code to produce your results, you are requested to upload your model. To do this, you can go to \"More->Team->Your Model\" and upload an archive of your code. It is necessary to zip everything as a single file. Please note that this upload link becomes unavailable after the deadline of stage 1, so you will need to upload it before the end of stage 1. \n \n**What should I upload?**\n \nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those. However, you should submit the code used to generate those files. You can typically select two submissions for final scoring, so don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each. \n \n**What happens to my pre-trained model?**\n \nYou only need to include a README file to indicate where you can download the pre-trained files from. For example, if you used vgg16 from keras, you don’t need to upload the weights file, you only need to indicate where you got it from: https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5\n \n**What if my submission is too big?**\n \nOur uploader will handle reasonably large files. If you still think your model will be too large, you can instead upload a checksum of your archive file (such as an md5 or sha hash). Note that if you do win, you will still have to upload it. A common reason for folder being too big might be that you included too many of your non-code files. If you upload a checksum, finish in a prize spot, and are unable to subsequently provide an archive that matches the checksum, you will be removed from the competition.\n \n**What if I want my code to stay private if I don’t win**\n \nYou may upload an encrypted archive and provide the decryption key in the event you win and wish to claim a prize. Alternatively, you can use a checksum, as described above.\n \n**What happens if I want to change something in my code in the second stage?**\n \nWe expect you may need to make some “non scientific” alterations, such as changes to path names, in order to create your submissions for the second stage. You are allowed to re-train your model (including the stage one data), but your code should not change. You should not be doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long as it is fully automated.\n \n**Can I upload multiple times?**\n \nYou may upload as many model files as you wish. Kaggle only keeps the latest upload. Make sure your last upload is the right one!\n \n**Will the number of participants change in the second stage? Will I get a medal? Will I get points?**\n \nYes, the number of participants will be smaller since some people won't submit in the second stage. For the purposes of medals and points calculations, we will use the number of participants in the first stage to calculate points and medals.\n", "votes": 15 }, { "id": 192938, "postDate": "2017-06-15T06:28:10.860Z", "content": "I really wish you would put things like this in the official competition rules instead of burying them in the forums here. This was my first Kaggle competition and I had no idea that we needed to upload our code, let alone how to do so, and I only just found this post several hours after it is too late. Consequently, I am now ineligible for a prize -- not that I would have won one anyway, but I think for future competitions this should be improved. I especially find that More->Team->Your Model is an obscure location for the upload link. It would be more intuitive to group it under My Submissions where we upload everything else.
", "rawMarkdown": "I really wish you would put things like this in the official competition rules instead of burying them in the forums here. This was my first Kaggle competition and I had no idea that we needed to upload our code, let alone how to do so, and I only just found this post several hours after it is too late. Consequently, I am now ineligible for a prize -- not that I would have won one anyway, but I think for future competitions this should be improved. I especially find that **More->Team->Your Model** is an obscure location for the upload link. It would be more intuitive to group it under **My Submissions** where we upload everything else.", "votes": 7, "replies": [ { "id": 194014, "postDate": "2017-06-19T05:31:44.190Z", "content": "+1 to this comment.
", "rawMarkdown": "+1 to this comment." } ] }, { "id": 193174, "postDate": "2017-06-15T19:01:46.490Z", "content": "so as we see that stage2 has many simillar pictures as in train/stg1.. anything is going to be done about that?:( overfitting teams have clearly got this otherwise...
", "rawMarkdown": "so as we see that stage2 has many simillar pictures as in train/stg1.. anything is going to be done about that?:( overfitting teams have clearly got this otherwise...", "votes": 1 }, { "id": 191938, "postDate": "2017-06-12T06:51:14.447Z", "content": "Silly question - on the \"Manage Teams\" page, how do you delete uploaded files? I uploaded some files but I want to upload a newer version. I didn't see a way to delete files. Do I just give the files the same name as the ones already uploaded and then upload them again?
", "rawMarkdown": "Silly question - on the \"Manage Teams\" page, how do you delete uploaded files? I uploaded some files but I want to upload a newer version. I didn't see a way to delete files. Do I just give the files the same name as the ones already uploaded and then upload them again?", "votes": 1, "replies": [ { "id": 192303, "postDate": "2017-06-13T05:30:44.183Z", "content": "See the second to last question:
\n\n\n", "rawMarkdown": "See the second to last question:\n\n> Can I upload multiple times?\n> \n> You may upload as many model files as you wish. Kaggle only keeps the\n> latest upload. Make sure your last upload is the right one!\n\n", "votes": 1 } ] }, { "id": 189750, "postDate": "2017-06-06T19:47:08.693Z", "content": "Can I upload multiple times?
\n \nYou may upload as many model files as you wish. Kaggle only keeps the\n latest upload. Make sure your last upload is the right one!
\n
Since the first deadline is tomorrow, I would just like to confirm that the only mandatory step is accepting the competition rules. Other than that, action is only needed if we are teaming up, correct?
", "rawMarkdown": "Since the first deadline is tomorrow, I would just like to confirm that the only mandatory step is accepting the competition rules. Other than that, action is only needed if we are teaming up, correct?", "votes": 1, "replies": [ { "id": 189790, "postDate": "2017-06-06T20:58:59.500Z", "content": "Good question. When you ask, I'm wondering the same. It says we should upload our model within 1st stage of the competition. \"Please note that this upload link becomes unavailable after the deadline of stage 1....\" \nDoes stage 1 end by tomorrow?
", "rawMarkdown": "Good question. When you ask, I'm wondering the same. It says we should upload our model within 1st stage of the competition. *\"Please note that this upload link becomes unavailable after the deadline of stage 1....\"* \nDoes stage 1 end by tomorrow?" }, { "id": 189793, "postDate": "2017-06-06T21:04:34.517Z", "content": "According the Timeline:
\n\n\n", "rawMarkdown": "According the Timeline: \n> June 14, 2017 - Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.", "votes": 2 }, { "id": 189804, "postDate": "2017-06-06T21:26:43.020Z", "content": "June 14, 2017 - Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.
\n
Thank you! Then I maybe have time to retrain my model without the horrible leak I've introduced in my setup... I'll start re-training now!
", "rawMarkdown": "Thank you! Then I maybe have time to retrain my model without the horrible leak I've introduced in my setup... I'll start re-training now!" }, { "id": 189838, "postDate": "2017-06-06T23:07:45.337Z", "content": "So have we confirmed then that no action needs to be taken today?
", "rawMarkdown": "So have we confirmed then that no action needs to be taken today?" } ] }, { "id": 183270, "postDate": "2017-05-17T16:01:29.230Z", "content": "17% of train images are of type 1, 53% are of type 2 and 30% are of type 3. Stage 1 test set images have the same distribution. How is the distribution of stage 2 test set?
", "rawMarkdown": "17% of train images are of type 1, 53% are of type 2 and 30% are of type 3. Stage 1 test set images have the same distribution. How is the distribution of stage 2 test set?", "votes": 1 }, { "id": 182660, "postDate": "2017-05-15T07:08:24.423Z", "content": "Will the stage 1 test set labels be revealed when we enter stage 2?
", "rawMarkdown": "Will the stage 1 test set labels be revealed when we enter stage 2?", "votes": 1, "replies": [ { "id": 184190, "postDate": "2017-05-20T21:01:01.807Z", "content": "I have the same question
", "rawMarkdown": "I have the same question" }, { "id": 184191, "postDate": "2017-05-20T21:17:13.333Z", "content": "Yes
", "rawMarkdown": "Yes", "votes": 1 }, { "id": 185253, "postDate": "2017-05-24T17:23:08.187Z", "content": "Hi Wendy,
\n\nHow will uploading a checksum work if we have to adapt the code to include new directories after the Stage 1 labels are released? Would you consider releasing the stage 1 labels before the model upload deadline so that we can incorporate this and have everything debugged, etc?
", "rawMarkdown": "Hi Wendy,\n\nHow will uploading a checksum work if we have to adapt the code to include new directories after the Stage 1 labels are released? Would you consider releasing the stage 1 labels before the model upload deadline so that we can incorporate this and have everything debugged, etc? ", "votes": 1 }, { "id": 185481, "postDate": "2017-05-25T06:28:00.420Z", "content": "It wouldn't work very well - but I think if you win, you can submit two versions of your code, one with each checksum version, to prove that you didn't change your code.
", "rawMarkdown": "It wouldn't work very well - but I think if you win, you can submit two versions of your code, one with each checksum version, to prove that you didn't change your code. " }, { "id": 190696, "postDate": "2017-06-08T08:28:28.637Z", "content": "In what format will the labels be released ?
\n\nIntegrated into the current train/Type_x directories, or rather a csv with file - label mapping ?
", "rawMarkdown": "In what format will the labels be released ?\n\n\nIntegrated into the current train/Type_x directories, or rather a csv with file - label mapping ?" }, { "id": 191481, "postDate": "2017-06-10T15:06:41.343Z", "content": "Hi @Wendi, i'm sorry but this is not clear to me, should we include the code with the training of the stage1 test data in the files that we have to load before the delivery of their labels? Or are these the non-scientific changes mentioned in the faqs?
", "rawMarkdown": "Hi @Wendi, i'm sorry but this is not clear to me, should we include the code with the training of the stage1 test data in the files that we have to load before the delivery of their labels? Or are these the non-scientific changes mentioned in the faqs?" }, { "id": 191503, "postDate": "2017-06-10T16:30:13.063Z", "content": "It will be a CSV.
", "rawMarkdown": "It will be a CSV. ", "votes": -1 }, { "id": 191506, "postDate": "2017-06-10T16:31:48.557Z", "content": "@sergejc, I'm not very clear on what you meant, but if you're unsure, it's best to include it.
", "rawMarkdown": "@sergejc, I'm not very clear on what you meant, but if you're unsure, it's best to include it. ", "votes": -1 } ] }, { "id": 195027, "postDate": "2017-06-22T16:32:47.237Z", "content": "Hi Wendy,
\n\nI am wondering, would it be possible to release the stage 2 labels after the competition hosts have finalized the private leaderboard? This competition was a great learning experience but I don't think the research and insights gained have to stop with the end of the competition. Personally, I would like the stage 2 labels for verifying hypotheses about our submissions. Another way would be to allow us to still submit entries that got scored but not included in the standings.
", "rawMarkdown": "Hi Wendy,\n\nI am wondering, would it be possible to release the stage 2 labels after the competition hosts have finalized the private leaderboard? This competition was a great learning experience but I don't think the research and insights gained have to stop with the end of the competition. Personally, I would like the stage 2 labels for verifying hypotheses about our submissions. Another way would be to allow us to still submit entries that got scored but not included in the standings.", "votes": 2, "replies": [ { "id": 195073, "postDate": "2017-06-22T17:54:23.497Z", "content": "Hi partner! I think you can make submissions even after the competition end and see where you would have scored. In that way you can actually pull out the answers you are looking for. (by leaderboard mining) If you intend to make a further study on this, I think the stage two leaderboard can be perfect for final verification of a model.
", "rawMarkdown": "Hi partner! I think you can make submissions even after the competition end and see where you would have scored. In that way you can actually pull out the answers you are looking for. (by leaderboard mining) If you intend to make a further study on this, I think the stage two leaderboard can be perfect for final verification of a model." }, { "id": 195074, "postDate": "2017-06-22T17:58:03.733Z", "content": "Ah good! That's exactly what I was looking for :)
", "rawMarkdown": "Ah good! That's exactly what I was looking for :)" }, { "id": 195084, "postDate": "2017-06-22T18:03:58.013Z", "content": "Or even check with another dataset : \nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34670
", "rawMarkdown": "Or even check with another dataset : \nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34670" }, { "id": 195138, "postDate": "2017-06-22T20:16:47.077Z", "content": "Great idea! I'd also be interested in knowing if the scoring feature of the stage 2 data will be up indefinitely? @Wendy
", "rawMarkdown": "Great idea! I'd also be interested in knowing if the scoring feature of the stage 2 data will be up indefinitely? @Wendy" } ] }, { "id": 182517, "postDate": "2017-05-14T10:01:49.770Z", "content": "The current data set have some very heterogeneous images, especially the 'green' ones. Can you please confirm that the stage_2 dataset will be similar in that matter ?
", "rawMarkdown": "The current data set have some very heterogeneous images, especially the 'green' ones. Can you please confirm that the stage_2 dataset will be similar in that matter ? ", "votes": 2, "replies": [ { "id": 188341, "postDate": "2017-06-02T17:36:32.610Z", "content": "In the stage1 test set, there are 0 green-filter images, but the training set has a number of them. Should we expect that stage2 test set will or will not contain green-filter images?
", "rawMarkdown": "In the stage1 test set, there are 0 green-filter images, but the training set has a number of them. Should we expect that stage2 test set will or will not contain green-filter images?" }, { "id": 190808, "postDate": "2017-06-08T14:41:18.987Z", "content": "@Wendy Kan - we would really appreciate an answer here too.
\n\nThere is 0 green-filter images in the test set and very little in the train set (6 I believe). However there are a lot (~600, which is 10%) in additional ones. Could you please confirm this will stand in the stage 2 dataset ?
", "rawMarkdown": "@Wendy Kan - we would really appreciate an answer here too. \n\nThere is 0 green-filter images in the test set and very little in the train set (6 I believe). However there are a lot (~600, which is 10%) in additional ones. Could you please confirm this will stand in the stage 2 dataset ?", "votes": 2 } ] }, { "id": 643453, "postDate": "2019-10-07T14:20:13.870Z", "content": "
where is \"More->Team->Your Model\" ????
", "rawMarkdown": "\n\nwhere is \"More->Team->Your Model\" ????" }, { "id": 195230, "postDate": "2017-06-23T01:45:23.057Z", "content": "@Wendy Kan
\n Hi Wendy,
\n Why my submission and the information of Intel competition is missing? Yesterday my information of Intel competition I can see.
Hi Wendy,\nI've posted 1781 100% duplicates here: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34788#193645\nPlease exclude them from scoring
", "rawMarkdown": "Hi Wendy,\nI've posted 1781 100% duplicates here: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34788#193645\nPlease exclude them from scoring" }, { "id": 193591, "postDate": "2017-06-17T01:27:53.653Z", "content": "@Wendy Kan If I retrain my model including stage 1 test data in the training set, the previous model file (HDF5 file generated by keras) will be changed. How can I upload new HDF5 file (model submission is closed)? Do I need to upload it?
", "rawMarkdown": "@Wendy Kan If I retrain my model including stage 1 test data in the training set, the previous model file (HDF5 file generated by keras) will be changed. How can I upload new HDF5 file (model submission is closed)? Do I need to upload it?" }, { "id": 192855, "postDate": "2017-06-15T00:25:38.973Z", "content": "Hi, can I upload the additional file now? I uploaded one file about my model details, but forgot to add in code. Can I do it now
", "rawMarkdown": "Hi, can I upload the additional file now? I uploaded one file about my model details, but forgot to add in code. Can I do it now" }, { "id": 192088, "postDate": "2017-06-12T16:52:43.983Z", "content": "Hi Wendy, there is an open discussion going on here (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34441) and it would be helpful to have some official input. Basically, we are wondering how close is close enough in terms of reproducibility. Keras/tensorflow has some non-determinisms that exist even after fixing random seeds, although the deviation that these remaining stochastic elements cause is very minor. This is a fairly important detail to discuss because many of the competitors are using the Keras/tensorflow combination somewhere in their solution stack.
", "rawMarkdown": "Hi Wendy, there is an open discussion going on here (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34441) and it would be helpful to have some official input. Basically, we are wondering how close is close enough in terms of reproducibility. Keras/tensorflow has some non-determinisms that exist even after fixing random seeds, although the deviation that these remaining stochastic elements cause is very minor. This is a fairly important detail to discuss because many of the competitors are using the Keras/tensorflow combination somewhere in their solution stack." }, { "id": 190499, "postDate": "2017-06-07T19:57:02.553Z", "content": "Hey , \nwe are a group who are working on the competition for practice , will the submission and evaluation be available after the end of the competition ?? regardless of the leader board ranking for sure .\nThanks
", "rawMarkdown": "Hey , \nwe are a group who are working on the competition for practice , will the submission and evaluation be available after the end of the competition ?? regardless of the leader board ranking for sure .\nThanks " }, { "id": 189839, "postDate": "2017-06-06T23:16:20.220Z", "content": "After the end of the second stage, will the second stage test set labels be released?
", "rawMarkdown": "After the end of the second stage, will the second stage test set labels be released?", "replies": [ { "id": 190509, "postDate": "2017-06-07T20:10:16.920Z", "content": "No, we won't release 2nd stage labels.
", "rawMarkdown": "No, we won't release 2nd stage labels. " } ] }, { "id": 187851, "postDate": "2017-06-01T10:30:19.563Z", "content": "Hi Wendy,
\n\nCan you provide a ballpark estimate of the size of stage2 test set? So we can plan for computing time accordingly... thanks
", "rawMarkdown": "Hi Wendy,\n\nCan you provide a ballpark estimate of the size of stage2 test set? So we can plan for computing time accordingly... thanks", "replies": [ { "id": 188356, "postDate": "2017-06-02T18:06:15.340Z", "content": "Yes, it's going to be around 3500 images (~12GB).
", "rawMarkdown": "Yes, it's going to be around 3500 images (~12GB). ", "votes": 2 }, { "id": 188803, "postDate": "2017-06-04T04:47:58.867Z", "content": "Hi Wendy,
\n\nWould it be possible to release the 12GB file a few days before stage2 begin ? So that those who have slow connection can start downloading it.
\n\nYou can password protect the file, then release the password upon stage2 begin.
\n\nThanks!
", "rawMarkdown": "Hi Wendy,\n\nWould it be possible to release the 12GB file a few days before stage2 begin ? So that those who have slow connection can start downloading it.\n\nYou can password protect the file, then release the password upon stage2 begin.\n\nThanks!" }, { "id": 189338, "postDate": "2017-06-05T20:20:48.200Z", "content": "Good idea, I'll do that. Thanks for suggesting.
\n\nWill 10 days (instead of 7) before close be enough time?
", "rawMarkdown": "Good idea, I'll do that. Thanks for suggesting. \n\nWill 10 days (instead of 7) before close be enough time?", "votes": 2 }, { "id": 189461, "postDate": "2017-06-06T05:35:49.913Z", "content": "Thanks ! Yes. More than enough for me :)
", "rawMarkdown": "Thanks ! Yes. More than enough for me :)" }, { "id": 191764, "postDate": "2017-06-11T17:47:31.973Z", "content": "So can we expect the stage2 files in coming day?
", "rawMarkdown": "So can we expect the stage2 files in coming day?", "votes": 1 }, { "id": 192059, "postDate": "2017-06-12T15:19:08.480Z", "content": "According to Kaggle time, 9 days to go and still no sign of 2nd stage data...
", "rawMarkdown": "According to Kaggle time, 9 days to go and still no sign of 2nd stage data..." }, { "id": 192307, "postDate": "2017-06-13T05:32:19.410Z", "content": "It was uploaded today (UTC). You should have had access to it for a few hours now.
", "rawMarkdown": "It was uploaded today (UTC). You should have had access to it for a few hours now. " }, { "id": 192359, "postDate": "2017-06-13T08:36:24.303Z", "content": "will stage 1 labels be released soon ?
", "rawMarkdown": "will stage 1 labels be released soon ?" }, { "id": 192405, "postDate": "2017-06-13T12:52:50.660Z", "content": "Can you provide sample_submission file for stage 2 so we can create a consistent sumbission file?
", "rawMarkdown": "Can you provide sample_submission file for stage 2 so we can create a consistent sumbission file?" }, { "id": 192453, "postDate": "2017-06-13T15:35:59.827Z", "content": "@Sayed AbdEl-Aziz, @Oleggrinch,\nStage 2 officially starts this Thursday, June 15. By then I'll provide both stage 1 labels and stage 2 sample submission file. The file now is just a making it easier for people who don't have fast internet to start their downloads early.
", "rawMarkdown": "@Sayed AbdEl-Aziz, @Oleggrinch,\nStage 2 officially starts this Thursday, June 15. By then I'll provide both stage 1 labels and stage 2 sample submission file. The file now is just a making it easier for people who don't have fast internet to start their downloads early. \n", "votes": 1 } ] }, { "id": 186210, "postDate": "2017-05-27T03:46:34.490Z", "content": "Hi Wendy, I noticed there were some updated labels. Can those labels be updated in the colfax server as well?
\n\nI believe 80 should be type 3, and 968 and 1120 type 1.
\n\nThanks!
\n\n$ ssh colfax\n######################################################################\n# Welcome to Colfax Cluster!\n######################################################################\n# If you are here for the Intel/MobileODT Kaggle contest,\n# (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening)\n# you can find the data-set at /data/kaggle on both the login node\n# and the compute nodes.\n# Note: If you are using the \"additional\" data, please use the ones\n# found in /data/kaggle_3.27/additional\n#\n# We have a dedicated forum page for the contest at:\n# https://colfaxresearch.com/discussion/forum/kaggle-contest-2017/\n#\n# Pre-compiled Machine Learning Frameworks are available in the /opt/ directory\n#\n# Colfax Research Team\n######################################################################\nLast login: Fri May 26 20:43:37 2017 from 10.5.0.7\n[u3737@c001 ~]$ cd /data/kaggle/train/\n[u3737@c001 train]$ find . -name 80.jpg\n./Type_2/80.jpg\n[u3737@c001 train]$ find . -name 968.jpg\n./Type_3/968.jpg\n[u3737@c001 train]$ find . -name 1120.jpg\n./Type_3/1120.jpg\n[u3737@c001 train]$\nDoes the code have to absolutely reproduce the models/submissions? Just so I know if I should spend time on fixing keras random seeds...
", "rawMarkdown": "Does the code have to absolutely reproduce the models/submissions? Just so I know if I should spend time on fixing keras random seeds...", "replies": [ { "id": 185804, "postDate": "2017-05-25T23:39:57.503Z", "content": "Theoretically yes, practically, it depends on the situation. It is possible for you to be rejected from claiming the prize if your code doesn't generate exactly the outcome because of random seed.
", "rawMarkdown": "Theoretically yes, practically, it depends on the situation. It is possible for you to be rejected from claiming the prize if your code doesn't generate exactly the outcome because of random seed. ", "votes": 1 }, { "id": 189111, "postDate": "2017-06-05T09:47:32.510Z", "content": "@Wendy\nIf someone is rejected from claiming the prize due to random seed, will his/her name be removed from the final leaderboard?
", "rawMarkdown": "@Wendy\nIf someone is rejected from claiming the prize due to random seed, will his/her name be removed from the final leaderboard?", "votes": 1 } ] }, { "id": 180641, "postDate": "2017-05-06T09:12:35.163Z", "content": "Thanks Wendy , This is awesome , it's good to see basic stage two definitions as they will probably apply to many competitions in the future as well ...very clear and much appreciated!
", "rawMarkdown": "Thanks Wendy , This is awesome , it's good to see basic stage two definitions as they will probably apply to many competitions in the future as well ...very clear and much appreciated!" }, { "id": 193804, "postDate": "2017-06-18T06:03:57.357Z", "rawMarkdown": "", "isDeleted": true }, { "id": 192873, "postDate": "2017-06-15T02:25:21.307Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 192878, "postDate": "2017-06-15T02:44:09.373Z", "content": "Yes, that is non-scientific.
", "rawMarkdown": "Yes, that is non-scientific. " } ] }, { "id": 180646, "postDate": "2017-05-06T09:39:29.463Z", "content": "Precise and clear. Thanks a lot!!!!! :)
", "rawMarkdown": "Precise and clear. Thanks a lot!!!!! :)" } ], "comments": [ { "id": 192938, "author_name": "Brett", "author_url": "", "post_date": "2017-06-15T06:28:10.860000", "content": "I really wish you would put things like this in the official competition rules instead of burying them in the forums here. This was my first Kaggle competition and I had no idea that we needed to upload our code, let alone how to do so, and I only just found this post several hours after it is too late. Consequently, I am now ineligible for a prize -- not that I would have won one anyway, but I think for future competitions this should be improved. I especially find that More->Team->Your Model is an obscure location for the upload link. It would be more intuitive to group it under My Submissions where we upload everything else.
", "votes": 7, "replies": [ { "id": 194014, "author_name": "aravind gowrisankar", "author_url": "", "post_date": "2017-06-19T05:31:44.190000", "content": "+1 to this comment.
", "votes": 0, "replies": [] } ] }, { "id": 193174, "author_name": "raddar", "author_url": "", "post_date": "2017-06-15T19:01:46.490000", "content": "so as we see that stage2 has many simillar pictures as in train/stg1.. anything is going to be done about that?:( overfitting teams have clearly got this otherwise...
", "votes": 1, "replies": [] }, { "id": 191938, "author_name": "PhillipChin", "author_url": "", "post_date": "2017-06-12T06:51:14.447000", "content": "Silly question - on the \"Manage Teams\" page, how do you delete uploaded files? I uploaded some files but I want to upload a newer version. I didn't see a way to delete files. Do I just give the files the same name as the ones already uploaded and then upload them again?
", "votes": 1, "replies": [ { "id": 192303, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-13T05:30:44.183000", "content": "See the second to last question:
\n\n\n", "votes": 1, "replies": [] } ] }, { "id": 189750, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-06T19:47:08.693000", "content": "Can I upload multiple times?
\n \nYou may upload as many model files as you wish. Kaggle only keeps the\n latest upload. Make sure your last upload is the right one!
\n
Since the first deadline is tomorrow, I would just like to confirm that the only mandatory step is accepting the competition rules. Other than that, action is only needed if we are teaming up, correct?
", "votes": 1, "replies": [ { "id": 189790, "author_name": "Øystein Schønning-Johansen", "author_url": "", "post_date": "2017-06-06T20:58:59.500000", "content": "Good question. When you ask, I'm wondering the same. It says we should upload our model within 1st stage of the competition. \"Please note that this upload link becomes unavailable after the deadline of stage 1....\" \nDoes stage 1 end by tomorrow?
", "votes": 0, "replies": [] }, { "id": 189793, "author_name": "vfdev", "author_url": "", "post_date": "2017-06-06T21:04:34.517000", "content": "According the Timeline:
\n\n\n", "votes": 2, "replies": [] }, { "id": 189804, "author_name": "Øystein Schønning-Johansen", "author_url": "", "post_date": "2017-06-06T21:26:43.020000", "content": "June 14, 2017 - Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.
\n
Thank you! Then I maybe have time to retrain my model without the horrible leak I've introduced in my setup... I'll start re-training now!
", "votes": 0, "replies": [] }, { "id": 189838, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-06T23:07:45.337000", "content": "So have we confirmed then that no action needs to be taken today?
", "votes": 0, "replies": [] } ] }, { "id": 183270, "author_name": "Buus", "author_url": "", "post_date": "2017-05-17T16:01:29.230000", "content": "17% of train images are of type 1, 53% are of type 2 and 30% are of type 3. Stage 1 test set images have the same distribution. How is the distribution of stage 2 test set?
", "votes": 1, "replies": [] }, { "id": 182660, "author_name": "Yusaku Sako", "author_url": "", "post_date": "2017-05-15T07:08:24.423000", "content": "Will the stage 1 test set labels be revealed when we enter stage 2?
", "votes": 1, "replies": [ { "id": 184190, "author_name": "ZFTurbo", "author_url": "", "post_date": "2017-05-20T21:01:01.807000", "content": "I have the same question
", "votes": 0, "replies": [] }, { "id": 184191, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-05-20T21:17:13.333000", "content": "Yes
", "votes": 1, "replies": [] }, { "id": 185253, "author_name": "fergusoci", "author_url": "", "post_date": "2017-05-24T17:23:08.187000", "content": "Hi Wendy,
\n\nHow will uploading a checksum work if we have to adapt the code to include new directories after the Stage 1 labels are released? Would you consider releasing the stage 1 labels before the model upload deadline so that we can incorporate this and have everything debugged, etc?
", "votes": 1, "replies": [] }, { "id": 185481, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-05-25T06:28:00.420000", "content": "It wouldn't work very well - but I think if you win, you can submit two versions of your code, one with each checksum version, to prove that you didn't change your code.
", "votes": 0, "replies": [] }, { "id": 190696, "author_name": "Alchemist", "author_url": "", "post_date": "2017-06-08T08:28:28.637000", "content": "In what format will the labels be released ?
\n\nIntegrated into the current train/Type_x directories, or rather a csv with file - label mapping ?
", "votes": 0, "replies": [] }, { "id": 191481, "author_name": "sergio callarelli", "author_url": "", "post_date": "2017-06-10T15:06:41.343000", "content": "Hi @Wendi, i'm sorry but this is not clear to me, should we include the code with the training of the stage1 test data in the files that we have to load before the delivery of their labels? Or are these the non-scientific changes mentioned in the faqs?
", "votes": 0, "replies": [] }, { "id": 191503, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-10T16:30:13.063000", "content": "It will be a CSV.
", "votes": -1, "replies": [] }, { "id": 191506, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-10T16:31:48.557000", "content": "@sergejc, I'm not very clear on what you meant, but if you're unsure, it's best to include it.
", "votes": -1, "replies": [] } ] }, { "id": 195027, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-22T16:32:47.237000", "content": "Hi Wendy,
\n\nI am wondering, would it be possible to release the stage 2 labels after the competition hosts have finalized the private leaderboard? This competition was a great learning experience but I don't think the research and insights gained have to stop with the end of the competition. Personally, I would like the stage 2 labels for verifying hypotheses about our submissions. Another way would be to allow us to still submit entries that got scored but not included in the standings.
", "votes": 2, "replies": [ { "id": 195073, "author_name": "Øystein Schønning-Johansen", "author_url": "", "post_date": "2017-06-22T17:54:23.497000", "content": "Hi partner! I think you can make submissions even after the competition end and see where you would have scored. In that way you can actually pull out the answers you are looking for. (by leaderboard mining) If you intend to make a further study on this, I think the stage two leaderboard can be perfect for final verification of a model.
", "votes": 0, "replies": [] }, { "id": 195074, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-22T17:58:03.733000", "content": "Ah good! That's exactly what I was looking for :)
", "votes": 0, "replies": [] }, { "id": 195084, "author_name": "vfdev", "author_url": "", "post_date": "2017-06-22T18:03:58.013000", "content": "Or even check with another dataset : \nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34670
", "votes": 0, "replies": [] }, { "id": 195138, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-22T20:16:47.077000", "content": "Great idea! I'd also be interested in knowing if the scoring feature of the stage 2 data will be up indefinitely? @Wendy
", "votes": 0, "replies": [] } ] }, { "id": 182517, "author_name": "Alchemist", "author_url": "", "post_date": "2017-05-14T10:01:49.770000", "content": "The current data set have some very heterogeneous images, especially the 'green' ones. Can you please confirm that the stage_2 dataset will be similar in that matter ?
", "votes": 2, "replies": [ { "id": 188341, "author_name": "Yusaku Sako", "author_url": "", "post_date": "2017-06-02T17:36:32.610000", "content": "In the stage1 test set, there are 0 green-filter images, but the training set has a number of them. Should we expect that stage2 test set will or will not contain green-filter images?
", "votes": 0, "replies": [] }, { "id": 190808, "author_name": "Alchemist", "author_url": "", "post_date": "2017-06-08T14:41:18.987000", "content": "@Wendy Kan - we would really appreciate an answer here too.
\n\nThere is 0 green-filter images in the test set and very little in the train set (6 I believe). However there are a lot (~600, which is 10%) in additional ones. Could you please confirm this will stand in the stage 2 dataset ?
", "votes": 2, "replies": [] } ] }, { "id": 643453, "author_name": "pupil3", "author_url": "", "post_date": "2019-10-07T14:20:13.870000", "content": "
where is \"More->Team->Your Model\" ????
", "votes": 0, "replies": [] }, { "id": 195230, "author_name": "Tyan", "author_url": "", "post_date": "2017-06-23T01:45:23.057000", "content": "@Wendy Kan
\n Hi Wendy,
\n Why my submission and the information of Intel competition is missing? Yesterday my information of Intel competition I can see.
Hi Wendy,\nI've posted 1781 100% duplicates here: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34788#193645\nPlease exclude them from scoring
", "votes": 0, "replies": [] }, { "id": 193591, "author_name": "Rakib Hyder", "author_url": "", "post_date": "2017-06-17T01:27:53.653000", "content": "@Wendy Kan If I retrain my model including stage 1 test data in the training set, the previous model file (HDF5 file generated by keras) will be changed. How can I upload new HDF5 file (model submission is closed)? Do I need to upload it?
", "votes": 0, "replies": [] }, { "id": 192855, "author_name": "VikramanK", "author_url": "", "post_date": "2017-06-15T00:25:38.973000", "content": "Hi, can I upload the additional file now? I uploaded one file about my model details, but forgot to add in code. Can I do it now
", "votes": 0, "replies": [] }, { "id": 192088, "author_name": "gkericks", "author_url": "", "post_date": "2017-06-12T16:52:43.983000", "content": "Hi Wendy, there is an open discussion going on here (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34441) and it would be helpful to have some official input. Basically, we are wondering how close is close enough in terms of reproducibility. Keras/tensorflow has some non-determinisms that exist even after fixing random seeds, although the deviation that these remaining stochastic elements cause is very minor. This is a fairly important detail to discuss because many of the competitors are using the Keras/tensorflow combination somewhere in their solution stack.
", "votes": 0, "replies": [] }, { "id": 190499, "author_name": "Sayed AbdEl-Aziz", "author_url": "", "post_date": "2017-06-07T19:57:02.553000", "content": "Hey , \nwe are a group who are working on the competition for practice , will the submission and evaluation be available after the end of the competition ?? regardless of the leader board ranking for sure .\nThanks
", "votes": 0, "replies": [] }, { "id": 189839, "author_name": "Marwa Said", "author_url": "", "post_date": "2017-06-06T23:16:20.220000", "content": "After the end of the second stage, will the second stage test set labels be released?
", "votes": 0, "replies": [ { "id": 190509, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-07T20:10:16.920000", "content": "No, we won't release 2nd stage labels.
", "votes": 0, "replies": [] } ] }, { "id": 187851, "author_name": "kubilai", "author_url": "", "post_date": "2017-06-01T10:30:19.563000", "content": "Hi Wendy,
\n\nCan you provide a ballpark estimate of the size of stage2 test set? So we can plan for computing time accordingly... thanks
", "votes": 0, "replies": [ { "id": 188356, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-02T18:06:15.340000", "content": "Yes, it's going to be around 3500 images (~12GB).
", "votes": 2, "replies": [] }, { "id": 188803, "author_name": "Scotty", "author_url": "", "post_date": "2017-06-04T04:47:58.867000", "content": "Hi Wendy,
\n\nWould it be possible to release the 12GB file a few days before stage2 begin ? So that those who have slow connection can start downloading it.
\n\nYou can password protect the file, then release the password upon stage2 begin.
\n\nThanks!
", "votes": 0, "replies": [] }, { "id": 189338, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-05T20:20:48.200000", "content": "Good idea, I'll do that. Thanks for suggesting.
\n\nWill 10 days (instead of 7) before close be enough time?
", "votes": 2, "replies": [] }, { "id": 189461, "author_name": "Scotty", "author_url": "", "post_date": "2017-06-06T05:35:49.913000", "content": "Thanks ! Yes. More than enough for me :)
", "votes": 0, "replies": [] }, { "id": 191764, "author_name": "raddar", "author_url": "", "post_date": "2017-06-11T17:47:31.973000", "content": "So can we expect the stage2 files in coming day?
", "votes": 1, "replies": [] }, { "id": 192059, "author_name": "Wojtek Rosinski", "author_url": "", "post_date": "2017-06-12T15:19:08.480000", "content": "According to Kaggle time, 9 days to go and still no sign of 2nd stage data...
", "votes": 0, "replies": [] }, { "id": 192307, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-13T05:32:19.410000", "content": "It was uploaded today (UTC). You should have had access to it for a few hours now.
", "votes": 0, "replies": [] }, { "id": 192359, "author_name": "Sayed AbdEl-Aziz", "author_url": "", "post_date": "2017-06-13T08:36:24.303000", "content": "will stage 1 labels be released soon ?
", "votes": 0, "replies": [] }, { "id": 192405, "author_name": "Oleggrinch", "author_url": "", "post_date": "2017-06-13T12:52:50.660000", "content": "Can you provide sample_submission file for stage 2 so we can create a consistent sumbission file?
", "votes": 0, "replies": [] }, { "id": 192453, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-13T15:35:59.827000", "content": "@Sayed AbdEl-Aziz, @Oleggrinch,\nStage 2 officially starts this Thursday, June 15. By then I'll provide both stage 1 labels and stage 2 sample submission file. The file now is just a making it easier for people who don't have fast internet to start their downloads early.
", "votes": 1, "replies": [] } ] }, { "id": 186210, "author_name": "Carlos Aguayo", "author_url": "", "post_date": "2017-05-27T03:46:34.490000", "content": "Hi Wendy, I noticed there were some updated labels. Can those labels be updated in the colfax server as well?
\n\nI believe 80 should be type 3, and 968 and 1120 type 1.
\n\nThanks!
\n\n$ ssh colfax\n######################################################################\n# Welcome to Colfax Cluster!\n######################################################################\n# If you are here for the Intel/MobileODT Kaggle contest,\n# (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening)\n# you can find the data-set at /data/kaggle on both the login node\n# and the compute nodes.\n# Note: If you are using the \"additional\" data, please use the ones\n# found in /data/kaggle_3.27/additional\n#\n# We have a dedicated forum page for the contest at:\n# https://colfaxresearch.com/discussion/forum/kaggle-contest-2017/\n#\n# Pre-compiled Machine Learning Frameworks are available in the /opt/ directory\n#\n# Colfax Research Team\n######################################################################\nLast login: Fri May 26 20:43:37 2017 from 10.5.0.7\n[u3737@c001 ~]$ cd /data/kaggle/train/\n[u3737@c001 train]$ find . -name 80.jpg\n./Type_2/80.jpg\n[u3737@c001 train]$ find . -name 968.jpg\n./Type_3/968.jpg\n[u3737@c001 train]$ find . -name 1120.jpg\n./Type_3/1120.jpg\n[u3737@c001 train]$\nDoes the code have to absolutely reproduce the models/submissions? Just so I know if I should spend time on fixing keras random seeds...
", "votes": 0, "replies": [ { "id": 185804, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-05-25T23:39:57.503000", "content": "Theoretically yes, practically, it depends on the situation. It is possible for you to be rejected from claiming the prize if your code doesn't generate exactly the outcome because of random seed.
", "votes": 1, "replies": [] }, { "id": 189111, "author_name": "chicm", "author_url": "", "post_date": "2017-06-05T09:47:32.510000", "content": "@Wendy\nIf someone is rejected from claiming the prize due to random seed, will his/her name be removed from the final leaderboard?
", "votes": 1, "replies": [] } ] }, { "id": 180641, "author_name": "Lea", "author_url": "", "post_date": "2017-05-06T09:12:35.163000", "content": "Thanks Wendy , This is awesome , it's good to see basic stage two definitions as they will probably apply to many competitions in the future as well ...very clear and much appreciated!
", "votes": 0, "replies": [] }, { "id": 193804, "author_name": "", "author_url": "", "post_date": "2017-06-18T06:03:57.357000", "content": "", "votes": 0, "replies": [] }, { "id": 192873, "author_name": "", "author_url": "", "post_date": "2017-06-15T02:25:21.307000", "content": "", "votes": 0, "replies": [ { "id": 192878, "author_name": "Wendy Kan", "author_url": "", "post_date": "2017-06-15T02:44:09.373000", "content": "Yes, that is non-scientific.
", "votes": 0, "replies": [] } ] }, { "id": 180646, "author_name": "SarthakYadav", "author_url": "", "post_date": "2017-05-06T09:39:29.463000", "content": "Precise and clear. Thanks a lot!!!!! :)
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "180543": "We'd like to remind you what will happen in the second stage of this competition:\n \n \n**Why second stage? Why so much trouble?**\n \nThe spirit of having a second stage is to prevent hand labeling and leaderboard probing of the test data. In order to achieve this, we ask that you upload your source code, including the correct parameters that you used for generating your submission files. This is for you to prove that you have written automated code to create your final submission(s). These \"models\" that you submit may be examined by Kaggle and the competition host to determine your eligibility to win the competition and claim prizes.\n \n**If I don't have a chance to win, should I upload my model?**\n \nYes. You never know where your team will place on private leaderboard, so it’s worth uploading.\n \n**Will I still be on the leaderboard if I don't submit in the second stage?**\n \nYou will not. You will need to make a submission in the correct format in the second stage to remain on the leaderboard. \n \n**Will I still be on the leaderboard if I don't upload a model but do submit in the second stage?**\n \nYes. However, if you do not upload a model and finish in a prize position, your team will be removed from the competition standings entirely!\n \n**How do I upload my model?** \n \nTo ensure that you did write code to produce your results, you are requested to upload your model. To do this, you can go to \"More->Team->Your Model\" and upload an archive of your code. It is necessary to zip everything as a single file. Please note that this upload link becomes unavailable after the deadline of stage 1, so you will need to upload it before the end of stage 1. \n \n**What should I upload?**\n \nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those. However, you should submit the code used to generate those files. You can typically select two submissions for final scoring, so don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each. \n \n**What happens to my pre-trained model?**\n \nYou only need to include a README file to indicate where you can download the pre-trained files from. For example, if you used vgg16 from keras, you don’t need to upload the weights file, you only need to indicate where you got it from: https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5\n \n**What if my submission is too big?**\n \nOur uploader will handle reasonably large files. If you still think your model will be too large, you can instead upload a checksum of your archive file (such as an md5 or sha hash). Note that if you do win, you will still have to upload it. A common reason for folder being too big might be that you included too many of your non-code files. If you upload a checksum, finish in a prize spot, and are unable to subsequently provide an archive that matches the checksum, you will be removed from the competition.\n \n**What if I want my code to stay private if I don’t win**\n \nYou may upload an encrypted archive and provide the decryption key in the event you win and wish to claim a prize. Alternatively, you can use a checksum, as described above.\n \n**What happens if I want to change something in my code in the second stage?**\n \nWe expect you may need to make some “non scientific” alterations, such as changes to path names, in order to create your submissions for the second stage. You are allowed to re-train your model (including the stage one data), but your code should not change. You should not be doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long as it is fully automated.\n \n**Can I upload multiple times?**\n \nYou may upload as many model files as you wish. Kaggle only keeps the latest upload. Make sure your last upload is the right one!\n \n**Will the number of participants change in the second stage? Will I get a medal? Will I get points?**\n \nYes, the number of participants will be smaller since some people won't submit in the second stage. For the purposes of medals and points calculations, we will use the number of participants in the first stage to calculate points and medals.\n", "192938": "I really wish you would put things like this in the official competition rules instead of burying them in the forums here. This was my first Kaggle competition and I had no idea that we needed to upload our code, let alone how to do so, and I only just found this post several hours after it is too late. Consequently, I am now ineligible for a prize -- not that I would have won one anyway, but I think for future competitions this should be improved. I especially find that **More->Team->Your Model** is an obscure location for the upload link. It would be more intuitive to group it under **My Submissions** where we upload everything else.", "193174": "so as we see that stage2 has many simillar pictures as in train/stg1.. anything is going to be done about that?:( overfitting teams have clearly got this otherwise...", "191938": "Silly question - on the \"Manage Teams\" page, how do you delete uploaded files? I uploaded some files but I want to upload a newer version. I didn't see a way to delete files. Do I just give the files the same name as the ones already uploaded and then upload them again?", "189750": "Since the first deadline is tomorrow, I would just like to confirm that the only mandatory step is accepting the competition rules. Other than that, action is only needed if we are teaming up, correct?", "183270": "17% of train images are of type 1, 53% are of type 2 and 30% are of type 3. Stage 1 test set images have the same distribution. How is the distribution of stage 2 test set?", "182660": "Will the stage 1 test set labels be revealed when we enter stage 2?", "195027": "Hi Wendy,\n\nI am wondering, would it be possible to release the stage 2 labels after the competition hosts have finalized the private leaderboard? This competition was a great learning experience but I don't think the research and insights gained have to stop with the end of the competition. Personally, I would like the stage 2 labels for verifying hypotheses about our submissions. Another way would be to allow us to still submit entries that got scored but not included in the standings.", "182517": "The current data set have some very heterogeneous images, especially the 'green' ones. Can you please confirm that the stage_2 dataset will be similar in that matter ? ", "643453": "\n\nwhere is \"More->Team->Your Model\" ????", "195230": "@Wendy Kan \n Hi Wendy, \n Why my submission and the information of Intel competition is missing? Yesterday my information of Intel competition I can see.", "193654": "Hi Wendy,\nI've posted 1781 100% duplicates here: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34788#193645\nPlease exclude them from scoring", "193591": "@Wendy Kan If I retrain my model including stage 1 test data in the training set, the previous model file (HDF5 file generated by keras) will be changed. How can I upload new HDF5 file (model submission is closed)? Do I need to upload it?", "192855": "Hi, can I upload the additional file now? I uploaded one file about my model details, but forgot to add in code. Can I do it now", "192088": "Hi Wendy, there is an open discussion going on here (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/34441) and it would be helpful to have some official input. Basically, we are wondering how close is close enough in terms of reproducibility. Keras/tensorflow has some non-determinisms that exist even after fixing random seeds, although the deviation that these remaining stochastic elements cause is very minor. This is a fairly important detail to discuss because many of the competitors are using the Keras/tensorflow combination somewhere in their solution stack.", "190499": "Hey , \nwe are a group who are working on the competition for practice , will the submission and evaluation be available after the end of the competition ?? regardless of the leader board ranking for sure .\nThanks ", "189839": "After the end of the second stage, will the second stage test set labels be released?", "187851": "Hi Wendy,\n\nCan you provide a ballpark estimate of the size of stage2 test set? So we can plan for computing time accordingly... thanks", "186210": "Hi Wendy, I noticed there were some updated labels. Can those labels be updated in the colfax server as well?\n\nI believe 80 should be type 3, and 968 and 1120 type 1.\n\nThanks!\n\n $ ssh colfax\n ######################################################################\n # Welcome to Colfax Cluster!\n ######################################################################\n # If you are here for the Intel/MobileODT Kaggle contest,\n # (https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening)\n # you can find the data-set at /data/kaggle on both the login node\n # and the compute nodes.\n # Note: If you are using the \"additional\" data, please use the ones\n # found in /data/kaggle_3.27/additional\n #\n # We have a dedicated forum page for the contest at:\n # https://colfaxresearch.com/discussion/forum/kaggle-contest-2017/\n #\n # Pre-compiled Machine Learning Frameworks are available in the /opt/ directory\n #\n # Colfax Research Team\n ######################################################################\n Last login: Fri May 26 20:43:37 2017 from 10.5.0.7\n [u3737@c001 ~]$ cd /data/kaggle/train/\n [u3737@c001 train]$ find . -name 80.jpg\n ./Type_2/80.jpg\n [u3737@c001 train]$ find . -name 968.jpg\n ./Type_3/968.jpg\n [u3737@c001 train]$ find . -name 1120.jpg\n ./Type_3/1120.jpg\n [u3737@c001 train]$", "185165": "Does the code have to absolutely reproduce the models/submissions? Just so I know if I should spend time on fixing keras random seeds...", "180641": "Thanks Wendy , This is awesome , it's good to see basic stage two definitions as they will probably apply to many competitions in the future as well ...very clear and much appreciated!", "193804": "", "192873": "", "180646": "Precise and clear. Thanks a lot!!!!! :)" } }