Hi all,
\n\nPlease post your questions/bugs/feedback about Intel Tools in this thread. They'll be happy to answer questions here.
", "messages": [ { "id": "168253", "postDate": "03/16/2017 17:53:58", "content": "Hi all,
\n\nPlease post your questions/bugs/feedback about Intel Tools in this thread. They'll be happy to answer questions here.
", "rawMarkdown": "Hi all,\n\nPlease post your questions/bugs/feedback about Intel Tools in this thread. They'll be happy to answer questions here.", "votes": null }, { "id": "169563", "postDate": "03/21/2017 14:38:22", "content": "I've installed the Deep Learning Training Tool on a google cloud linux server (running Ubuntu 14.04). Uploading data through the HTML user interface is taking a long time and fails to complete. Is there another way to upload data to the training tool besides using the HTML user interface?
\n\nThanks
", "rawMarkdown": "I've installed the Deep Learning Training Tool on a google cloud linux server (running Ubuntu 14.04). Uploading data through the HTML user interface is taking a long time and fails to complete. Is there another way to upload data to the training tool besides using the HTML user interface?\n\nThanks", "votes": null }, { "id": "169575", "postDate": "03/21/2017 15:42:50", "content": "Hi Guy,
\n\nUnfortunately not, at this point it needs to go through the interface because it builds a small database for usage in this process. I good mechanism to help you out though is do not upload the the large images to your instance. Pre-process the images to a smaller size before uploading.
", "rawMarkdown": "Hi Guy,\n\nUnfortunately not, at this point it needs to go through the interface because it builds a small database for usage in this process. I good mechanism to help you out though is do not upload the the large images to your instance. Pre-process the images to a smaller size before uploading.", "votes": null }, { "id": "169792", "postDate": "03/22/2017 16:31:41", "content": "I installed the Intel Training Tool locally on my mac. I am not clear whether I need to install anything else to be able to train models (such as a mac version of deployment tool). Does the training tool include Caffe optimized for Intel?
", "rawMarkdown": "I installed the Intel Training Tool locally on my mac. I am not clear whether I need to install anything else to be able to train models (such as a mac version of deployment tool). Does the training tool include Caffe optimized for Intel?", "votes": null }, { "id": "170283", "postDate": "03/24/2017 21:12:34", "content": "Yes, training tool includes Intel-optimized version of the DL framework - Caffe. Installer downloaded all required packages/software. So, it should be ready to use at this point.
", "rawMarkdown": "Yes, training tool includes Intel-optimized version of the DL framework - Caffe. Installer downloaded all required packages/software. So, it should be ready to use at this point.", "votes": null }, { "id": "170455", "postDate": "03/25/2017 23:01:26", "content": "I have installed Intel SDK in MAC OSX. However the when creating Datasets the Dataset info only indicates very few items were loaded. I tried the set of data from the competition as well as the example for mnist. All of them load well with the expected size, however when creating the dataset no more than 13 items are loaded.
\n\nI appreciate your help
", "rawMarkdown": "I have installed Intel SDK in MAC OSX. However the when creating Datasets the Dataset info only indicates very few items were loaded. I tried the set of data from the competition as well as the example for mnist. All of them load well with the expected size, however when creating the dataset no more than 13 items are loaded.\n\nI appreciate your help", "votes": null }, { "id": "170814", "postDate": "03/27/2017 17:32:25", "content": "Do you have images in the separate folders (where folder name is the label)?
\n\nFor instance, like this:
\n\n> - mnist\n> - 1\n> 0001.png\n> 0002.png\n> ....\n> - 2\n> 0101.png\n> 0102.png\n> ...\n> - 3 ...\n> \n> - 9 ...\n> - 0 ...\nHi, it shows that the license will expire by 4/30/2017. How can I extend the license?
", "rawMarkdown": "Hi, it shows that the license will expire by 4/30/2017. How can I extend the license?", "votes": null }, { "id": "171062", "postDate": "03/28/2017 14:23:43", "content": "Hi Vadim\nThanks for your answer. Yes I do have the folders as per the data one folder per each type \"Type_1\", \"Type_2\" and \"Type_3\", I also tried with the example with the mnist files for handwriting recognition.\nI am using the original files name in both cases.\nThe upload files size seems ok however when creating the data set it ends at o seconds and only a few items (5 or 13) are indicated. \nThanks for your help.
\n\nPS:The training app seems very useful for the Kaggle competition, I hope I am able to use it.
", "rawMarkdown": "Hi Vadim\nThanks for your answer. Yes I do have the folders as per the data one folder per each type \"Type_1\", \"Type_2\" and \"Type_3\", I also tried with the example with the mnist files for handwriting recognition.\nI am using the original files name in both cases.\nThe upload files size seems ok however when creating the data set it ends at o seconds and only a few items (5 or 13) are indicated. \nThanks for your help.\n\nPS:The training app seems very useful for the Kaggle competition, I hope I am able to use it.", "votes": null }, { "id": "171287", "postDate": "03/29/2017 09:48:18", "content": "This SDK does not support large dataset upload. And the solution for this is using \"docker cp\" in Mac OS and unzip the uploaded zip in Linux OS.
", "rawMarkdown": "This SDK does not support large dataset upload. And the solution for this is using \"docker cp\" in Mac OS and unzip the uploaded zip in Linux OS.", "votes": null }, { "id": "171428", "postDate": "03/29/2017 23:22:52", "content": "Hi Intel Team,
\n\nI just want to point you guy to my prev post asking for help when installing the tool:
\n\nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30287
\n\nWe haven't been able yet to debug the process.\nAny help would be appreciated.
\n\nRegards,\nAndrei
", "rawMarkdown": "Hi Intel Team,\n\nI just want to point you guy to my prev post asking for help when installing the tool:\n\nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30287\n\nWe haven't been able yet to debug the process.\nAny help would be appreciated.\n\nRegards,\nAndrei", "votes": null }, { "id": "174981", "postDate": "04/13/2017 10:53:04", "content": "I am currently downloading the intel SDK package, havent tried it yet\nHowever, my curiosity leads me to ask, what is the goal of intel SDK? Isn't using GPU faster to compute these type of calculations?
", "rawMarkdown": "I am currently downloading the intel SDK package, havent tried it yet\nHowever, my curiosity leads me to ask, what is the goal of intel SDK? Isn't using GPU faster to compute these type of calculations?", "votes": null }, { "id": "175060", "postDate": "04/13/2017 16:27:22", "content": "Did anyone compare Intel Distribution for Python with regular Anaconda + Gohlke (http://www.lfd.uci.edu/~gohlke/pythonlibs/) libraries linked with Intel MKL? Any benefits of going the Intel route?
", "rawMarkdown": "Did anyone compare Intel Distribution for Python with regular Anaconda + Gohlke (http://www.lfd.uci.edu/~gohlke/pythonlibs/) libraries linked with Intel MKL? Any benefits of going the Intel route?", "votes": null }, { "id": "175116", "postDate": "04/13/2017 20:52:05", "content": "Not sure if this is considered part of the SDK but I would like to bring to your attention Possible security vulnerability on colfax
", "rawMarkdown": "Not sure if this is considered part of the SDK but I would like to bring to your attention [Possible security vulnerability on colfax][1]\n\n [1]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/31592/", "votes": null }, { "id": "175117", "postDate": "04/13/2017 20:52:19", "content": "", "rawMarkdown": "", "votes": null }, { "id": "175858", "postDate": "04/17/2017 22:10:05", "content": "Hi,
\n\nI'm just beginning to figure out the Intel SDK setup. Here are some super basic questions.
\n\nThanks,\nSrinivas
", "rawMarkdown": "Hi,\n\nI'm just beginning to figure out the Intel SDK setup. Here are some super basic questions.\n\n1. Will I be able to install and run it on a windows OS machine instead of a linux/MAC machine ? \n2. Will I be able to access the datasets provided without physically downloading them ? \n3. I do not see a step in the tutorial where the user defines a feature set. Is there a standard feature set that the tool automatically uses ? Is there a way for users to specify specific features to consider ?\n\nThanks,\nSrinivas", "votes": null }, { "id": "176962", "postDate": "04/22/2017 21:04:54", "content": "Hi Team,
\n\nSame issue on my end. I've set up an AWS Ubuntu server, successfully ran through the install steps; however, when trying to access the GUI via the web, my Chrome browser times out....
\n\n\n\nI've tried different variations of the port number (31000) on the end, but still no luck....any suggestions recommended.
", "rawMarkdown": "Hi Team,\n\nSame issue on my end. I've set up an AWS Ubuntu server, successfully ran through the install steps; however, when trying to access the GUI via the web, my Chrome browser times out....\n\nhttp://34.209.228.140:31000/\n\nI've tried different variations of the port number (31000) on the end, but still no luck....any suggestions recommended.", "votes": null }, { "id": "177135", "postDate": "04/23/2017 20:23:47", "content": "I have successfully logged into my colfax server but unable to establish connect with SDK training tool. What should be done?
\n\nScreenshot: https://drive.google.com/file/d/0B2_R1dzouqHBbFRlcHZxMDBpSnc/view?usp=sharing
", "rawMarkdown": "I have successfully logged into my colfax server but unable to establish connect with SDK training tool. What should be done?\n\nScreenshot: https://drive.google.com/file/d/0B2_R1dzouqHBbFRlcHZxMDBpSnc/view?usp=sharing", "votes": null }, { "id": "177152", "postDate": "04/24/2017 00:22:14", "content": "Hi Guy - apologies in advance as I'm responding to your question w/ a question of my own; but wanted to know if you ever had any initial problems accessing the HTML user interface?
", "rawMarkdown": "Hi Guy - apologies in advance as I'm responding to your question w/ a question of my own; but wanted to know if you ever had any initial problems accessing the HTML user interface?", "votes": null }, { "id": "177765", "postDate": "04/25/2017 20:54:11", "content": "Hello,\nSome of you have reported that you are unable to connect to the Intel Deep Learning SDK interface. Please download and try the new release \"1.0 Beta 2\" and let us know if the problem persists.
", "rawMarkdown": "Hello,\nSome of you have reported that you are unable to connect to the Intel Deep Learning SDK interface. Please download and try the new release \"1.0 Beta 2\" and let us know if the problem persists.", "votes": null }, { "id": "179524", "postDate": "05/01/2017 20:56:38", "content": "Problem fixed (thanks Dominik and Noha!). Just needed to make sure the port that the tool uses was accessible on my remote server machine.
\n\nHowever, now I'm having challenges loading the dataset! Tried .zip, .rar and folder/subfolder configuration is correct. It goes all the way to 100% loading and right at 100% it fails indicating:
\n\nUpload failed. \nThe compressed folder doesn`t match any of of the supported structures
", "rawMarkdown": "Problem fixed (thanks Dominik and Noha!). Just needed to make sure the port that the tool uses was accessible on my remote server machine.\n\nHowever, now I'm having challenges loading the dataset! Tried .zip, .rar and folder/subfolder configuration is correct. It goes all the way to 100% loading and right at 100% it fails indicating:\n\n***Upload failed. \nThe compressed folder doesn`t match any of of the supported structures***", "votes": null }, { "id": "179644", "postDate": "05/02/2017 11:38:31", "content": "Hello Wendy,
\n\nI've fully deployed SDK and train few models, but can you point me to additional materials about SDK. Current questions:
\n\nExamples in Python notebook how to work with trained model files for predictions.
How SDK can work with few folds based on Dataset, for example, I want to split Dataset to 5 folds, train cross 5 models and predict mean results based on 5 models.
Thanks!
", "rawMarkdown": "Hello Wendy,\n\nI've fully deployed SDK and train few models, but can you point me to additional materials about SDK. Current questions:\n\n- Examples in Python notebook how to work with trained model files for predictions.\n\n- How SDK can work with few folds based on Dataset, for example, I want to split Dataset to 5 folds, train cross 5 models and predict mean results based on 5 models. \n\nThanks!", "votes": null }, { "id": "179676", "postDate": "05/02/2017 13:29:40", "content": "Bringing this here so there's no possibility of the administration team not being informed.
\n\nI have submitted my mined truth values as well just be clear that this isn't all talk.
\n\nI just want a statement from the management team on how they plan to deal with ground truth mining on the stage 1 LB, and if they intend to release the stage 1 ground truth with the stage 2 data. I have asked this a month ago when I first talked about the method and vulnerability of this competition, and again last week when the first 0 LB score was posted by a new account. If the management team isn't monitoring all posts in the forum, I urge them to do that as well.
\n\nI'll give this another day. Either contact me if you need more time, or just post your answer to those questions. If I don't hear back from the management team in 24h, I'll be posting the full stage 1 truth to level the playing field, and unfortunately this will have the side-effect of totally wrecking the stage 1 LB.
", "rawMarkdown": "[Bringing this here so there's no possibility of the administration team not being informed][1].\n\nI have submitted my mined truth values as well just be clear that this isn't all talk.\n\nI just want a statement from the management team on how they plan to deal with ground truth mining on the stage 1 LB, and if they intend to release the stage 1 ground truth with the stage 2 data. [I have asked this a month ago when I first talked about the method and vulnerability of this competition][2], and again last week when the first 0 LB score was posted by a new account. If the management team isn't monitoring all posts in the forum, I urge them to do that as well.\n\nI'll give this another day. Either contact me if you need more time, or just post your answer to those questions. If I don't hear back from the management team in 24h, I'll be posting the full stage 1 truth to level the playing field, and unfortunately this will have the side-effect of totally wrecking the stage 1 LB.\n\n\n [1]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/32213#178738\n [2]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30150#172109", "votes": null }, { "id": "179708", "postDate": "05/02/2017 15:58:40", "content": "I am just slightly behind you on LB. Because of hardware limitation, I can only submit so many results a day. Let's just say that I have little idea on how ground truth mining works, and I have never run a script in my life to mine the ground truth in any dataset. It seems unfair to me if the ground truth is posted, doesn't it?
", "rawMarkdown": "I am just slightly behind you on LB. Because of hardware limitation, I can only submit so many results a day. Let's just say that I have little idea on how ground truth mining works, and I have never run a script in my life to mine the ground truth in any dataset. It seems unfair to me if the ground truth is posted, doesn't it?", "votes": null }, { "id": "179711", "postDate": "05/02/2017 16:10:17", "content": "Someone who masters a technique oblique to the purpose of this challenge gets 512 extra training samples, so what is fairer? Should I keep this to myself and the other half-dozen teams who have access to this, or make this public and let everybody have the same tools to compete? The only thing at stake here is the stage 1 LB, which is already compromised.
\n\nIf the competition managements come forward and assumes the burden of keeping the LB clean and releasing the stage 1 truth with the stage 2 data, I'm content with holding onto it. I'm just talking about releasing it outright instead of releasing a script to generate the submissions like Oleg did in the DSB because it takes more than 2 weeks to fully extract the data, and the competition doesn't have all that much time left.
", "rawMarkdown": "Someone who masters a technique oblique to the purpose of this challenge gets 512 extra training samples, so what is fairer? Should I keep this to myself and the other half-dozen teams who have access to this, or make this public and let everybody have the same tools to compete? The only thing at stake here is the stage 1 LB, which is already compromised.\n\nIf the competition managements come forward and assumes the burden of keeping the LB clean and releasing the stage 1 truth with the stage 2 data, I'm content with holding onto it. I'm just talking about releasing it outright instead of releasing a script to generate the submissions like Oleg did in the DSB because it takes more than 2 weeks to fully extract the data, and the competition doesn't have all that much time left.", "votes": null }, { "id": "179717", "postDate": "05/02/2017 16:48:20", "content": "Ah I think I just got what you are advocating. To be fair, I think I would be benefited if the stage 1 result is released.
", "rawMarkdown": "Ah I think I just got what you are advocating. To be fair, I think I would be benefited if the stage 1 result is released.", "votes": null }, { "id": "179822", "postDate": "05/02/2017 22:32:16", "content": "Thank you for posting this again. We will release a stage 2 dataset, and will release stage 1's ground truth along with it.
", "rawMarkdown": "Thank you for posting this again. We will release a stage 2 dataset, and will release stage 1's ground truth along with it.", "votes": null }, { "id": "179825", "postDate": "05/02/2017 23:16:11", "content": "Thanks for the response. I'll hold on the data.
\n\nAs for our current #1 LB entry, I'd like also to hear about the policy you're going to follow regarding it. There's 3 possibilities on how it got posted:
\n\nHi, I could not install intel deep learning sdk on AWS. The installation log shows that it ran out of disk space.
\n\n5/13/2017, 11:41:50 AM info failed to register layer: Error processing tar file(exit status 1): write /usr/lib/x86_64-linux-gnu/libsqlite3.a: no space left on device \n5/13/2017, 11:41:50 AM info Error on or near line 812; exiting with status 1 \n5/13/2017, 11:41:50 AM info
\n5/13/2017, 11:41:50 AM error error: Error: Command failed: tools\\win\\plink.exe -i \"C:\\Users\\sunggeun\\Desktop\\ubuntu_key.ppk\" -load intel-training-tool-cloud -batch -t -ssh \"chmod +x dlsdk_install_scripts/*.sh && sudo -S -E -k dlsdk_install_scripts/install_training_tool.sh -stage 2 -type single -username 'ubuntu' -startport 8080 -selfsigncertificate '' \"
\n5/13/2017, 11:41:50 AM error docker pull completed with errors
Hi Wendy, I was looking to download the intel SDK and get this message when I click on the link in Kaggle contest.
\n\nError\nThank you for your interest in Intel® Deep Learning SDK Beta \nNOTICE: The current BETA program ended on 2017/04/30. The Intel® Deep Learning SDK Beta is no longer available.
\n\nHas the download been moved somewhere? \nWhen I go to Intel's website, I see there is Beta 2. \nhttps://software.intel.com/en-us/deep-learning-training-tool\nCan we use this tool for the contest?
\n\nThanks in advance.
", "rawMarkdown": "Hi Wendy, I was looking to download the intel SDK and get this message when I click on the link in Kaggle contest.\n\n*Error\nThank you for your interest in Intel® Deep Learning SDK Beta \nNOTICE: The current BETA program ended on 2017/04/30. The Intel® Deep Learning SDK Beta is no longer available.*\n\nHas the download been moved somewhere? \nWhen I go to Intel's website, I see there is Beta 2. \nhttps://software.intel.com/en-us/deep-learning-training-tool\nCan we use this tool for the contest?\n\nThanks in advance.", "votes": null }, { "id": "189228", "postDate": "06/05/2017 14:22:56", "content": "Hi iball41, \nDid you solve the upload problem? It is the same for me and I'm sure I'm using the folder/subfolder structure and have also tried .zip and .rar.\nThanks
", "rawMarkdown": "Hi iball41, \nDid you solve the upload problem? It is the same for me and I'm sure I'm using the folder/subfolder structure and have also tried .zip and .rar.\nThanks", "votes": null } ], "comments": [ { "id": 169563, "author_name": "zahaviguy", "author_url": "", "post_date": "03/21/2017 14:38:22", "content": "I've installed the Deep Learning Training Tool on a google cloud linux server (running Ubuntu 14.04). Uploading data through the HTML user interface is taking a long time and fails to complete. Is there another way to upload data to the training tool besides using the HTML user interface?
\n\nThanks
", "votes": null, "replies": [ { "id": 177152, "author_name": "iball41", "author_url": "", "post_date": "04/24/2017 00:22:14", "content": "Hi Guy - apologies in advance as I'm responding to your question w/ a question of my own; but wanted to know if you ever had any initial problems accessing the HTML user interface?
", "votes": null, "replies": [] } ] }, { "id": 169575, "author_name": "benodom", "author_url": "", "post_date": "03/21/2017 15:42:50", "content": "Hi Guy,
\n\nUnfortunately not, at this point it needs to go through the interface because it builds a small database for usage in this process. I good mechanism to help you out though is do not upload the the large images to your instance. Pre-process the images to a smaller size before uploading.
", "votes": null, "replies": [] }, { "id": 169792, "author_name": "salpert", "author_url": "", "post_date": "03/22/2017 16:31:41", "content": "I installed the Intel Training Tool locally on my mac. I am not clear whether I need to install anything else to be able to train models (such as a mac version of deployment tool). Does the training tool include Caffe optimized for Intel?
", "votes": null, "replies": [ { "id": 170283, "author_name": "vadimintel", "author_url": "", "post_date": "03/24/2017 21:12:34", "content": "Yes, training tool includes Intel-optimized version of the DL framework - Caffe. Installer downloaded all required packages/software. So, it should be ready to use at this point.
", "votes": null, "replies": [] } ] }, { "id": 170455, "author_name": "lrincones", "author_url": "", "post_date": "03/25/2017 23:01:26", "content": "I have installed Intel SDK in MAC OSX. However the when creating Datasets the Dataset info only indicates very few items were loaded. I tried the set of data from the competition as well as the example for mnist. All of them load well with the expected size, however when creating the dataset no more than 13 items are loaded.
\n\nI appreciate your help
", "votes": null, "replies": [ { "id": 170814, "author_name": "vadimintel", "author_url": "", "post_date": "03/27/2017 17:32:25", "content": "Do you have images in the separate folders (where folder name is the label)?
\n\nFor instance, like this:
\n\n> - mnist\n> - 1\n> 0001.png\n> 0002.png\n> ....\n> - 2\n> 0101.png\n> 0102.png\n> ...\n> - 3 ...\n> \n> - 9 ...\n> - 0 ...\nHi Vadim\nThanks for your answer. Yes I do have the folders as per the data one folder per each type \"Type_1\", \"Type_2\" and \"Type_3\", I also tried with the example with the mnist files for handwriting recognition.\nI am using the original files name in both cases.\nThe upload files size seems ok however when creating the data set it ends at o seconds and only a few items (5 or 13) are indicated. \nThanks for your help.
\n\nPS:The training app seems very useful for the Kaggle competition, I hope I am able to use it.
", "votes": null, "replies": [] } ] }, { "id": 170961, "author_name": "jiweiliu", "author_url": "", "post_date": "03/28/2017 05:57:33", "content": "Hi, it shows that the license will expire by 4/30/2017. How can I extend the license?
", "votes": null, "replies": [] }, { "id": 171287, "author_name": "gabriel0402", "author_url": "", "post_date": "03/29/2017 09:48:18", "content": "This SDK does not support large dataset upload. And the solution for this is using \"docker cp\" in Mac OS and unzip the uploaded zip in Linux OS.
", "votes": null, "replies": [] }, { "id": 171428, "author_name": "arukavina", "author_url": "", "post_date": "03/29/2017 23:22:52", "content": "Hi Intel Team,
\n\nI just want to point you guy to my prev post asking for help when installing the tool:
\n\nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30287
\n\nWe haven't been able yet to debug the process.\nAny help would be appreciated.
\n\nRegards,\nAndrei
", "votes": null, "replies": [] }, { "id": 174981, "author_name": "alberts123", "author_url": "", "post_date": "04/13/2017 10:53:04", "content": "I am currently downloading the intel SDK package, havent tried it yet\nHowever, my curiosity leads me to ask, what is the goal of intel SDK? Isn't using GPU faster to compute these type of calculations?
", "votes": null, "replies": [] }, { "id": 175060, "author_name": "sakvaua", "author_url": "", "post_date": "04/13/2017 16:27:22", "content": "Did anyone compare Intel Distribution for Python with regular Anaconda + Gohlke (http://www.lfd.uci.edu/~gohlke/pythonlibs/) libraries linked with Intel MKL? Any benefits of going the Intel route?
", "votes": null, "replies": [] }, { "id": 175116, "author_name": "", "author_url": "", "post_date": "04/13/2017 20:52:05", "content": "Not sure if this is considered part of the SDK but I would like to bring to your attention Possible security vulnerability on colfax
", "votes": null, "replies": [] }, { "id": 175117, "author_name": "", "author_url": "", "post_date": "04/13/2017 20:52:19", "content": "", "votes": null, "replies": [] }, { "id": 175858, "author_name": "srini1304", "author_url": "", "post_date": "04/17/2017 22:10:05", "content": "Hi,
\n\nI'm just beginning to figure out the Intel SDK setup. Here are some super basic questions.
\n\nThanks,\nSrinivas
", "votes": null, "replies": [] }, { "id": 176962, "author_name": "iball41", "author_url": "", "post_date": "04/22/2017 21:04:54", "content": "Hi Team,
\n\nSame issue on my end. I've set up an AWS Ubuntu server, successfully ran through the install steps; however, when trying to access the GUI via the web, my Chrome browser times out....
\n\n\n\nI've tried different variations of the port number (31000) on the end, but still no luck....any suggestions recommended.
", "votes": null, "replies": [ { "id": 179524, "author_name": "iball41", "author_url": "", "post_date": "05/01/2017 20:56:38", "content": "Problem fixed (thanks Dominik and Noha!). Just needed to make sure the port that the tool uses was accessible on my remote server machine.
\n\nHowever, now I'm having challenges loading the dataset! Tried .zip, .rar and folder/subfolder configuration is correct. It goes all the way to 100% loading and right at 100% it fails indicating:
\n\nUpload failed. \nThe compressed folder doesn`t match any of of the supported structures
", "votes": null, "replies": [] }, { "id": 189228, "author_name": "bongster21", "author_url": "", "post_date": "06/05/2017 14:22:56", "content": "Hi iball41, \nDid you solve the upload problem? It is the same for me and I'm sure I'm using the folder/subfolder structure and have also tried .zip and .rar.\nThanks
", "votes": null, "replies": [] } ] }, { "id": 177135, "author_name": "nikhilkhurana", "author_url": "", "post_date": "04/23/2017 20:23:47", "content": "I have successfully logged into my colfax server but unable to establish connect with SDK training tool. What should be done?
\n\nScreenshot: https://drive.google.com/file/d/0B2_R1dzouqHBbFRlcHZxMDBpSnc/view?usp=sharing
", "votes": null, "replies": [] }, { "id": 177765, "author_name": "meghanarao", "author_url": "", "post_date": "04/25/2017 20:54:11", "content": "Hello,\nSome of you have reported that you are unable to connect to the Intel Deep Learning SDK interface. Please download and try the new release \"1.0 Beta 2\" and let us know if the problem persists.
", "votes": null, "replies": [] }, { "id": 179644, "author_name": "szhitansky", "author_url": "", "post_date": "05/02/2017 11:38:31", "content": "Hello Wendy,
\n\nI've fully deployed SDK and train few models, but can you point me to additional materials about SDK. Current questions:
\n\nExamples in Python notebook how to work with trained model files for predictions.
How SDK can work with few folds based on Dataset, for example, I want to split Dataset to 5 folds, train cross 5 models and predict mean results based on 5 models.
Thanks!
", "votes": null, "replies": [] }, { "id": 179676, "author_name": "mumech", "author_url": "", "post_date": "05/02/2017 13:29:40", "content": "Bringing this here so there's no possibility of the administration team not being informed.
\n\nI have submitted my mined truth values as well just be clear that this isn't all talk.
\n\nI just want a statement from the management team on how they plan to deal with ground truth mining on the stage 1 LB, and if they intend to release the stage 1 ground truth with the stage 2 data. I have asked this a month ago when I first talked about the method and vulnerability of this competition, and again last week when the first 0 LB score was posted by a new account. If the management team isn't monitoring all posts in the forum, I urge them to do that as well.
\n\nI'll give this another day. Either contact me if you need more time, or just post your answer to those questions. If I don't hear back from the management team in 24h, I'll be posting the full stage 1 truth to level the playing field, and unfortunately this will have the side-effect of totally wrecking the stage 1 LB.
", "votes": null, "replies": [ { "id": 179708, "author_name": "chuijh", "author_url": "", "post_date": "05/02/2017 15:58:40", "content": "I am just slightly behind you on LB. Because of hardware limitation, I can only submit so many results a day. Let's just say that I have little idea on how ground truth mining works, and I have never run a script in my life to mine the ground truth in any dataset. It seems unfair to me if the ground truth is posted, doesn't it?
", "votes": null, "replies": [] }, { "id": 179711, "author_name": "mumech", "author_url": "", "post_date": "05/02/2017 16:10:17", "content": "Someone who masters a technique oblique to the purpose of this challenge gets 512 extra training samples, so what is fairer? Should I keep this to myself and the other half-dozen teams who have access to this, or make this public and let everybody have the same tools to compete? The only thing at stake here is the stage 1 LB, which is already compromised.
\n\nIf the competition managements come forward and assumes the burden of keeping the LB clean and releasing the stage 1 truth with the stage 2 data, I'm content with holding onto it. I'm just talking about releasing it outright instead of releasing a script to generate the submissions like Oleg did in the DSB because it takes more than 2 weeks to fully extract the data, and the competition doesn't have all that much time left.
", "votes": null, "replies": [] }, { "id": 179717, "author_name": "chuijh", "author_url": "", "post_date": "05/02/2017 16:48:20", "content": "Ah I think I just got what you are advocating. To be fair, I think I would be benefited if the stage 1 result is released.
", "votes": null, "replies": [] }, { "id": 179822, "author_name": "wendykan", "author_url": "", "post_date": "05/02/2017 22:32:16", "content": "Thank you for posting this again. We will release a stage 2 dataset, and will release stage 1's ground truth along with it.
", "votes": null, "replies": [] }, { "id": 179825, "author_name": "mumech", "author_url": "", "post_date": "05/02/2017 23:16:11", "content": "Thanks for the response. I'll hold on the data.
\n\nAs for our current #1 LB entry, I'd like also to hear about the policy you're going to follow regarding it. There's 3 possibilities on how it got posted:
\n\nHi, I could not install intel deep learning sdk on AWS. The installation log shows that it ran out of disk space.
\n\n5/13/2017, 11:41:50 AM info failed to register layer: Error processing tar file(exit status 1): write /usr/lib/x86_64-linux-gnu/libsqlite3.a: no space left on device \n5/13/2017, 11:41:50 AM info Error on or near line 812; exiting with status 1 \n5/13/2017, 11:41:50 AM info
\n5/13/2017, 11:41:50 AM error error: Error: Command failed: tools\\win\\plink.exe -i \"C:\\Users\\sunggeun\\Desktop\\ubuntu_key.ppk\" -load intel-training-tool-cloud -batch -t -ssh \"chmod +x dlsdk_install_scripts/*.sh && sudo -S -E -k dlsdk_install_scripts/install_training_tool.sh -stage 2 -type single -username 'ubuntu' -startport 8080 -selfsigncertificate '' \"
\n5/13/2017, 11:41:50 AM error docker pull completed with errors
Hi Wendy, I was looking to download the intel SDK and get this message when I click on the link in Kaggle contest.
\n\nError\nThank you for your interest in Intel® Deep Learning SDK Beta \nNOTICE: The current BETA program ended on 2017/04/30. The Intel® Deep Learning SDK Beta is no longer available.
\n\nHas the download been moved somewhere? \nWhen I go to Intel's website, I see there is Beta 2. \nhttps://software.intel.com/en-us/deep-learning-training-tool\nCan we use this tool for the contest?
\n\nThanks in advance.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "168253": "Hi all,\n\nPlease post your questions/bugs/feedback about Intel Tools in this thread. They'll be happy to answer questions here.", "169563": "I've installed the Deep Learning Training Tool on a google cloud linux server (running Ubuntu 14.04). Uploading data through the HTML user interface is taking a long time and fails to complete. Is there another way to upload data to the training tool besides using the HTML user interface?\n\nThanks", "169575": "Hi Guy,\n\nUnfortunately not, at this point it needs to go through the interface because it builds a small database for usage in this process. I good mechanism to help you out though is do not upload the the large images to your instance. Pre-process the images to a smaller size before uploading.", "169792": "I installed the Intel Training Tool locally on my mac. I am not clear whether I need to install anything else to be able to train models (such as a mac version of deployment tool). Does the training tool include Caffe optimized for Intel?", "170283": "Yes, training tool includes Intel-optimized version of the DL framework - Caffe. Installer downloaded all required packages/software. So, it should be ready to use at this point.", "170455": "I have installed Intel SDK in MAC OSX. However the when creating Datasets the Dataset info only indicates very few items were loaded. I tried the set of data from the competition as well as the example for mnist. All of them load well with the expected size, however when creating the dataset no more than 13 items are loaded.\n\nI appreciate your help", "170814": "Do you have images in the separate folders (where folder name is the label)?\n\nFor instance, like this:\n\n > - mnist\n > - 1\n > 0001.png\n > 0002.png\n > ....\n > - 2\n > 0101.png\n > 0102.png\n > ...\n > - 3 ...\n > \n > - 9 ...\n > - 0 ...", "170961": "Hi, it shows that the license will expire by 4/30/2017. How can I extend the license?", "171062": "Hi Vadim\nThanks for your answer. Yes I do have the folders as per the data one folder per each type \"Type_1\", \"Type_2\" and \"Type_3\", I also tried with the example with the mnist files for handwriting recognition.\nI am using the original files name in both cases.\nThe upload files size seems ok however when creating the data set it ends at o seconds and only a few items (5 or 13) are indicated. \nThanks for your help.\n\nPS:The training app seems very useful for the Kaggle competition, I hope I am able to use it.", "171287": "This SDK does not support large dataset upload. And the solution for this is using \"docker cp\" in Mac OS and unzip the uploaded zip in Linux OS.", "171428": "Hi Intel Team,\n\nI just want to point you guy to my prev post asking for help when installing the tool:\n\nhttps://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30287\n\nWe haven't been able yet to debug the process.\nAny help would be appreciated.\n\nRegards,\nAndrei", "174981": "I am currently downloading the intel SDK package, havent tried it yet\nHowever, my curiosity leads me to ask, what is the goal of intel SDK? Isn't using GPU faster to compute these type of calculations?", "175060": "Did anyone compare Intel Distribution for Python with regular Anaconda + Gohlke (http://www.lfd.uci.edu/~gohlke/pythonlibs/) libraries linked with Intel MKL? Any benefits of going the Intel route?", "175116": "Not sure if this is considered part of the SDK but I would like to bring to your attention [Possible security vulnerability on colfax][1]\n\n [1]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/31592/", "175117": "", "175858": "Hi,\n\nI'm just beginning to figure out the Intel SDK setup. Here are some super basic questions.\n\n1. Will I be able to install and run it on a windows OS machine instead of a linux/MAC machine ? \n2. Will I be able to access the datasets provided without physically downloading them ? \n3. I do not see a step in the tutorial where the user defines a feature set. Is there a standard feature set that the tool automatically uses ? Is there a way for users to specify specific features to consider ?\n\nThanks,\nSrinivas", "176962": "Hi Team,\n\nSame issue on my end. I've set up an AWS Ubuntu server, successfully ran through the install steps; however, when trying to access the GUI via the web, my Chrome browser times out....\n\nhttp://34.209.228.140:31000/\n\nI've tried different variations of the port number (31000) on the end, but still no luck....any suggestions recommended.", "177135": "I have successfully logged into my colfax server but unable to establish connect with SDK training tool. What should be done?\n\nScreenshot: https://drive.google.com/file/d/0B2_R1dzouqHBbFRlcHZxMDBpSnc/view?usp=sharing", "177152": "Hi Guy - apologies in advance as I'm responding to your question w/ a question of my own; but wanted to know if you ever had any initial problems accessing the HTML user interface?", "177765": "Hello,\nSome of you have reported that you are unable to connect to the Intel Deep Learning SDK interface. Please download and try the new release \"1.0 Beta 2\" and let us know if the problem persists.", "179524": "Problem fixed (thanks Dominik and Noha!). Just needed to make sure the port that the tool uses was accessible on my remote server machine.\n\nHowever, now I'm having challenges loading the dataset! Tried .zip, .rar and folder/subfolder configuration is correct. It goes all the way to 100% loading and right at 100% it fails indicating:\n\n***Upload failed. \nThe compressed folder doesn`t match any of of the supported structures***", "179644": "Hello Wendy,\n\nI've fully deployed SDK and train few models, but can you point me to additional materials about SDK. Current questions:\n\n- Examples in Python notebook how to work with trained model files for predictions.\n\n- How SDK can work with few folds based on Dataset, for example, I want to split Dataset to 5 folds, train cross 5 models and predict mean results based on 5 models. \n\nThanks!", "179676": "[Bringing this here so there's no possibility of the administration team not being informed][1].\n\nI have submitted my mined truth values as well just be clear that this isn't all talk.\n\nI just want a statement from the management team on how they plan to deal with ground truth mining on the stage 1 LB, and if they intend to release the stage 1 ground truth with the stage 2 data. [I have asked this a month ago when I first talked about the method and vulnerability of this competition][2], and again last week when the first 0 LB score was posted by a new account. If the management team isn't monitoring all posts in the forum, I urge them to do that as well.\n\nI'll give this another day. Either contact me if you need more time, or just post your answer to those questions. If I don't hear back from the management team in 24h, I'll be posting the full stage 1 truth to level the playing field, and unfortunately this will have the side-effect of totally wrecking the stage 1 LB.\n\n\n [1]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/32213#178738\n [2]: https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/discussion/30150#172109", "179708": "I am just slightly behind you on LB. Because of hardware limitation, I can only submit so many results a day. Let's just say that I have little idea on how ground truth mining works, and I have never run a script in my life to mine the ground truth in any dataset. It seems unfair to me if the ground truth is posted, doesn't it?", "179711": "Someone who masters a technique oblique to the purpose of this challenge gets 512 extra training samples, so what is fairer? Should I keep this to myself and the other half-dozen teams who have access to this, or make this public and let everybody have the same tools to compete? The only thing at stake here is the stage 1 LB, which is already compromised.\n\nIf the competition managements come forward and assumes the burden of keeping the LB clean and releasing the stage 1 truth with the stage 2 data, I'm content with holding onto it. I'm just talking about releasing it outright instead of releasing a script to generate the submissions like Oleg did in the DSB because it takes more than 2 weeks to fully extract the data, and the competition doesn't have all that much time left.", "179717": "Ah I think I just got what you are advocating. To be fair, I think I would be benefited if the stage 1 result is released.", "179822": "Thank you for posting this again. We will release a stage 2 dataset, and will release stage 1's ground truth along with it.", "179825": "Thanks for the response. I'll hold on the data.\n\nAs for our current #1 LB entry, I'd like also to hear about the policy you're going to follow regarding it. There's 3 possibilities on how it got posted:\n\n 1. The user guessed the whole truth in 3 tries (3 chances in 2^1024);\n 2. One of the other users mined the LB (there's about half a dozen users who have enough submissions to have finished it), and created an alternative account to post the submission, thus breaking the \"One account per participant\" rule;\n 3. One of the other users mined the LB, and sent the truth values to another user, outside of the forum, thus breaking the \"No private sharing outside teams\" rule;", "182424": "Hi, I could not install intel deep learning sdk on AWS. The installation log shows that it ran out of disk space.\n\n5/13/2017, 11:41:50 AM \t info \t failed to register layer: Error processing tar file(exit status 1): write /usr/lib/x86_64-linux-gnu/libsqlite3.a: no space left on device \n5/13/2017, 11:41:50 AM \t info \t Error on or near line 812; exiting with status 1 \n5/13/2017, 11:41:50 AM \t info \t \n5/13/2017, 11:41:50 AM \t error \t error: