New to machine learning and data science? No question is too basic or too simple. Feel free to start your own thread, or use this thread as a place to post any first-timer clarifying questions for the Kaggle community to help you with!
\n\nIf you would consider yourself a beginner but don't know where to get started, let other Kagglers help you take your first steps here!
\n\nNew to Kaggle? Take a look at a few videos our very own Dr. Rachael Tatman has put together to learn a bit more about site etiquette, Kaggle lingo, and how to enter a competition using Kaggle Notebooks.
\n\nReady to dive into this competition? Review the Overview Description which also provides links to docs and a getting started tutorial.
", "messages": [ { "id": 744833, "postDate": "2020-02-13T07:27:03.353Z", "content": "New to machine learning and data science? No question is too basic or too simple. Feel free to start your own thread, or use this thread as a place to post any first-timer clarifying questions for the Kaggle community to help you with!
\n\nIf you would consider yourself a beginner but don't know where to get started, let other Kagglers help you take your first steps here!
\n\nNew to Kaggle? Take a look at a few videos our very own Dr. Rachael Tatman has put together to learn a bit more about site etiquette, Kaggle lingo, and how to enter a competition using Kaggle Notebooks.
\n\nReady to dive into this competition? Review the Overview Description which also provides links to docs and a getting started tutorial.
", "rawMarkdown": "New to machine learning and data science? No question is too basic or too simple. Feel free to start your own thread, or use this thread as a place to post any first-timer clarifying questions for the Kaggle community to help you with! \n\nIf you would consider yourself a beginner but don't know where to get started, let other Kagglers help you take your first steps here!\n\nNew to Kaggle? Take a look at a few videos our very own Dr. Rachael Tatman has put together to learn a bit more about [site etiquette](https://www.youtube.com/watch?v=aIus8si_Et0), [Kaggle lingo](https://www.youtube.com/watch?v=sEJHyuWKd-s), and [how to enter a competition using Kaggle Notebooks](https://www.youtube.com/watch?&v=GJBOMWpLpTQ).\n\nReady to dive into this competition? Review the [Overview Description](https://www.kaggle.com/c/flower-classification-with-tpus/overview) which also provides links to docs and a getting started tutorial.", "votes": 8 }, { "id": 745062, "postDate": "2020-02-13T12:44:07Z", "content": "\n\n\nTPUs read training data exclusively from Google Cloud Storage.
\n
There is something I want to confirm.\nIs there a charge when using TPU in Notebook?
", "rawMarkdown": ">TPUs read training data exclusively from Google Cloud Storage.\n\nThere is something I want to confirm.\nIs there a charge when using TPU in Notebook?", "votes": 1, "replies": [ { "id": 745344, "postDate": "2020-02-13T18:19:08.167Z", "content": "No, no charge. Kaggle takes care of provisioning the TPU for you and of setting up a GCS bucket for the Kaggle dataset(s) you are using as well.
\n\nYou can get the GCS path of the Kaggle dataset you are using by writing:\n\ngcs_path = KaggleDatasets().get_gcs_path()\n
If you are reading from your own GCS bucket that you created on GCP (using you own account and billing), there can be egress charges as for any egress from a bucket.
\n\nTPU training from any public GCS bucket is supported, but if the bucket is not owned by Kaggle, the owner pays for egress.
", "rawMarkdown": "If you are reading from your own GCS bucket that you created on GCP (using you own account and billing), there can be egress charges as for any egress from a bucket.\n\nTPU training from any public GCS bucket is supported, but if the bucket is not owned by Kaggle, the owner pays for egress.", "votes": 3 }, { "id": 745573, "postDate": "2020-02-14T00:14:12.837Z", "content": "Thank you for your courteous reply.\nI feeled relieved.
", "rawMarkdown": "Thank you for your courteous reply.\nI feeled relieved." } ] }, { "id": 832420, "postDate": "2020-05-04T05:46:47.167Z", "content": "Hi, i just want to mention guide Submit Prediction is a bit confusing for Kaggle newcomers like me.
\n\nThe \"Step 1\" says you should click \"Submit\", but i couldn't find it (on Kernel type Notebook). After check all possible button, i found out that i must click \"Save Version\" and choose \"Save & Run All (Commit)\".
", "rawMarkdown": "Hi, i just want to mention guide [Submit Prediction](https://www.kaggle.com/c/flower-classification-with-tpus/submit) is a bit confusing for Kaggle newcomers like me.\n\nThe \"Step 1\" says you should click \"Submit\", but i couldn't find it (on Kernel type Notebook). After check all possible button, i found out that i must click \"Save Version\" and choose \"Save & Run All (Commit)\"." }, { "id": 828989, "postDate": "2020-05-01T12:29:16.070Z", "content": "Like Google Colab,Is there any functionality to connect runtime as TPU in kaggle Notebook?\nBecause After running this below code:\ntry:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver()
\n print('Running on TPU ', tpu.master())\nexcept ValueError:\n tpu = None
if tpu:\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n strategy = tf.distribute.get_strategy()
\n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)
\n\nI got output as below:
\n\nREPLICAS:1
\n\nAnd After fitting model I got error : \"Your Notebook tried to allocate more memory than its available .It has restarted\"
\n\nplease give solution
", "rawMarkdown": "Like Google Colab,Is there any functionality to connect runtime as TPU in kaggle Notebook?\nBecause After running this below code:\ntry:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver() \n print('Running on TPU ', tpu.master())\nexcept ValueError:\n tpu = None\n\nif tpu:\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n strategy = tf.distribute.get_strategy()\n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync) \n\n\nI got output as below:\n\nREPLICAS:1\n\nAnd After fitting model I got error : \"Your Notebook tried to allocate more memory than its available .It has restarted\"\n\nplease give solution" }, { "id": 824191, "postDate": "2020-04-28T08:03:18.337Z", "content": "I have a question whether or not the validation set can be used to train the final model in predicting the results?
", "rawMarkdown": "I have a question whether or not the validation set can be used to train the final model in predicting the results?" }, { "id": 815960, "postDate": "2020-04-22T01:20:27.053Z", "content": "Is it necessary to make modifications to BatchNorm or other layers to run correctly on TPUs? I noticed that the reference Efficientnet models in the tensorflow/tpu repo make some adjustments. If I understood correctly, it looked like the normalization gets calculated on a per replica basis and gets adjusted to calculate across replicas.
", "rawMarkdown": "Is it necessary to make modifications to BatchNorm or other layers to run correctly on TPUs? I noticed that the reference Efficientnet models in the [tensorflow/tpu repo](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/utils.py#L89) make some adjustments. If I understood correctly, it looked like the normalization gets calculated on a per replica basis and gets adjusted to calculate across replicas." }, { "id": 798430, "postDate": "2020-04-05T13:41:29.650Z", "content": "\n\nCan you guys please confirm that we have to use the tpu.estimator to PREDICT as well?
", "rawMarkdown": "@juliaelliott @mgornergoogle \n\nCan you guys please confirm that we have to use the **tpu.estimator** to PREDICT as well?", "replies": [ { "id": 801663, "postDate": "2020-04-08T17:13:12.423Z", "content": "You definitely can use tpu.estimator to predict, but it isn't required. If you do decide to use tpu.estimator please check out the documentation for some considerations.
No, please DO NOT use TPUEstimator. This is the old TPU API. May it rest in peace. Use Keras model.predict()
", "rawMarkdown": "No, please DO NOT use TPUEstimator. This is the old TPU API. May it rest in peace. Use Keras model.predict()", "votes": 2 }, { "id": 801682, "postDate": "2020-04-08T17:31:53.470Z", "content": "docs and sample here: https://www.kaggle.com/docs/tpu
", "rawMarkdown": "docs and sample here: https://www.kaggle.com/docs/tpu" }, { "id": 801692, "postDate": "2020-04-08T17:54:10.023Z", "content": "wonderful - thank you for the clarification!
", "rawMarkdown": "wonderful - thank you for the clarification!" }, { "id": 802112, "postDate": "2020-04-09T06:32:36.617Z", "content": "Thanks for clarification. Does Keras model.predict() use TPU for batch inference if the model is compiled under TPU strategy?
", "rawMarkdown": "Thanks for clarification. Does Keras model.predict() use TPU for batch inference if the model is compiled under TPU strategy?" } ] }, { "id": 787867, "postDate": "2020-03-27T06:37:42.553Z", "content": "Does it matter which image size I use to train the model? the training set has 4 different folders.
", "rawMarkdown": "Does it matter which image size I use to train the model? the training set has 4 different folders.", "replies": [ { "id": 788429, "postDate": "2020-03-27T16:58:22.163Z", "content": "Yes, usually larger images give a better final accuracy. They are much slower to train on though.\nAlso, when using pre-trained models, these tend to perform best on the image size they have been pre-trained on, even if they can support other sizes.
", "rawMarkdown": "Yes, usually larger images give a better final accuracy. They are much slower to train on though.\nAlso, when using pre-trained models, these tend to perform best on the image size they have been pre-trained on, even if they can support other sizes.", "votes": 1 } ] }, { "id": 787277, "postDate": "2020-03-26T17:04:59.613Z", "content": "Hi, I'm new to Kaggle and just wanted to clarify this statement in the description \"That code must be made public following the end of the competition.\" . Does it mean that I should make the kernel public?
", "rawMarkdown": "Hi, I'm new to Kaggle and just wanted to clarify this statement in the description \"That code must be made public following the end of the competition.\" . Does it mean that I should make the kernel public? ", "replies": [ { "id": 796506, "postDate": "2020-04-03T16:29:41.200Z", "content": "@jesudasdsouza To be eligible for the TPU leaderboard prizes, the prospective top 3 winners will need to share the code used to generate their submissions (using machine learning code, fully run on TPUs, including training) publicly. You can wait to do this after the competition closes (so as not to keep your approach private, until the end). To be eligible for the TPU Star prizes, a key aspect we will evaluate on are your upvoted, public contributions, so those notebooks/discussion posts should also be public.
", "rawMarkdown": "@jesudasdsouza To be eligible for the **TPU leaderboard prizes**, the prospective top 3 winners will need to share the code used to generate their submissions (using machine learning code, fully run on TPUs, including training) publicly. You can wait to do this after the competition closes (so as not to keep your approach private, until the end). To be eligible for the **TPU Star prizes**, a key aspect we will evaluate on are your *upvoted, public* contributions, so those notebooks/discussion posts should also be public.", "votes": 1 } ] }, { "id": 746625, "postDate": "2020-02-15T09:30:52.757Z", "content": "Amazing
", "rawMarkdown": "Amazing" }, { "id": 832177, "postDate": "2020-05-03T23:39:04.850Z", "rawMarkdown": "", "isDeleted": true }, { "id": 747532, "postDate": "2020-02-16T14:51:56.343Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 749537, "postDate": "2020-02-18T19:01:45.167Z", "content": "The training has to happen in the notebook. Using a notebook to read from a file and write the answers into the submission file will not make you eligible for prizes. That said, doing so is still fine. This is a friendly, \"playground\" competition. You are free to experiment with whatever you want to solve the problem at hand. Restrictions are only for the prize slots.
", "rawMarkdown": "The training has to happen in the notebook. Using a notebook to read from a file and write the answers into the submission file will not make you eligible for prizes. That said, doing so is still fine. This is a friendly, \"playground\" competition. You are free to experiment with whatever you want to solve the problem at hand. Restrictions are only for the prize slots. " }, { "id": 752384, "postDate": "2020-02-21T00:47:30.430Z", "content": "A correction to Martin's statement. Training does not have to be done in Kaggle notebooks. But per the Prizes page, it must have been trained on TPUs, although that can happen outside of Kaggle. We will be validating that is the case when reviewing the candidates for these awards. Inference and submission does have to be done in Kaggle notebooks.
\n\n> To be eligible for these prizes, your submission must generate predictions from machine learning code fully run on TPUs for both training and inference. That code must be made public following the end of the competition.
", "rawMarkdown": "A correction to Martin's statement. Training **does not** have to be done *in* Kaggle notebooks. But per the [Prizes page](https://www.kaggle.com/c/flower-classification-with-tpus/overview/prizes), it must have been trained on TPUs, although that can happen outside of Kaggle. We will be validating that is the case when reviewing the candidates for these awards. Inference and submission **does** have to be done *in* Kaggle notebooks.\n\n> To be eligible for these prizes, your submission must generate predictions from machine learning code fully run on TPUs for both training and inference. That code must be made public following the end of the competition." } ] }, { "id": 829302, "postDate": "2020-05-01T17:07:02.107Z", "content": "thanks
", "rawMarkdown": "thanks\n" }, { "id": 746533, "postDate": "2020-02-15T06:43:40.713Z", "content": "Thank you, is very useful
", "rawMarkdown": "Thank you, is very useful" }, { "id": 835872, "postDate": "2020-05-06T15:04:29.553Z", "content": "thank you
", "rawMarkdown": "thank you\n", "isDeleted": true } ], "comments": [ { "id": 745062, "author_name": "cocoa", "author_url": "", "post_date": "2020-02-13T12:44:07", "content": "\n\n\nTPUs read training data exclusively from Google Cloud Storage.
\n
There is something I want to confirm.\nIs there a charge when using TPU in Notebook?
", "votes": 1, "replies": [ { "id": 745344, "author_name": "Martin Görner", "author_url": "", "post_date": "2020-02-13T18:19:08.167000", "content": "No, no charge. Kaggle takes care of provisioning the TPU for you and of setting up a GCS bucket for the Kaggle dataset(s) you are using as well.
\n\nYou can get the GCS path of the Kaggle dataset you are using by writing:\n\ngcs_path = KaggleDatasets().get_gcs_path()\n
If you are reading from your own GCS bucket that you created on GCP (using you own account and billing), there can be egress charges as for any egress from a bucket.
\n\nTPU training from any public GCS bucket is supported, but if the bucket is not owned by Kaggle, the owner pays for egress.
", "votes": 3, "replies": [] }, { "id": 745573, "author_name": "cocoa", "author_url": "", "post_date": "2020-02-14T00:14:12.837000", "content": "Thank you for your courteous reply.\nI feeled relieved.
", "votes": 0, "replies": [] } ] }, { "id": 832420, "author_name": "Sandy Khosasi", "author_url": "", "post_date": "2020-05-04T05:46:47.167000", "content": "Hi, i just want to mention guide Submit Prediction is a bit confusing for Kaggle newcomers like me.
\n\nThe \"Step 1\" says you should click \"Submit\", but i couldn't find it (on Kernel type Notebook). After check all possible button, i found out that i must click \"Save Version\" and choose \"Save & Run All (Commit)\".
", "votes": 0, "replies": [] }, { "id": 828989, "author_name": "Siddharth Shah2601", "author_url": "", "post_date": "2020-05-01T12:29:16.070000", "content": "Like Google Colab,Is there any functionality to connect runtime as TPU in kaggle Notebook?\nBecause After running this below code:\ntry:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver()
\n print('Running on TPU ', tpu.master())\nexcept ValueError:\n tpu = None
if tpu:\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n strategy = tf.distribute.get_strategy()
\n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)
\n\nI got output as below:
\n\nREPLICAS:1
\n\nAnd After fitting model I got error : \"Your Notebook tried to allocate more memory than its available .It has restarted\"
\n\nplease give solution
", "votes": 0, "replies": [] }, { "id": 824191, "author_name": "ilearning99", "author_url": "", "post_date": "2020-04-28T08:03:18.337000", "content": "I have a question whether or not the validation set can be used to train the final model in predicting the results?
", "votes": 0, "replies": [] }, { "id": 815960, "author_name": "Caleb", "author_url": "", "post_date": "2020-04-22T01:20:27.053000", "content": "Is it necessary to make modifications to BatchNorm or other layers to run correctly on TPUs? I noticed that the reference Efficientnet models in the tensorflow/tpu repo make some adjustments. If I understood correctly, it looked like the normalization gets calculated on a per replica basis and gets adjusted to calculate across replicas.
", "votes": 0, "replies": [] }, { "id": 798430, "author_name": "Harsh Patel", "author_url": "", "post_date": "2020-04-05T13:41:29.650000", "content": "\n\nCan you guys please confirm that we have to use the tpu.estimator to PREDICT as well?
", "votes": 0, "replies": [ { "id": 801663, "author_name": "Jesse Mostipak", "author_url": "", "post_date": "2020-04-08T17:13:12.423000", "content": "You definitely can use tpu.estimator to predict, but it isn't required. If you do decide to use tpu.estimator please check out the documentation for some considerations.
No, please DO NOT use TPUEstimator. This is the old TPU API. May it rest in peace. Use Keras model.predict()
", "votes": 2, "replies": [] }, { "id": 801682, "author_name": "Martin Görner", "author_url": "", "post_date": "2020-04-08T17:31:53.470000", "content": "docs and sample here: https://www.kaggle.com/docs/tpu
", "votes": 0, "replies": [] }, { "id": 801692, "author_name": "Jesse Mostipak", "author_url": "", "post_date": "2020-04-08T17:54:10.023000", "content": "wonderful - thank you for the clarification!
", "votes": 0, "replies": [] }, { "id": 802112, "author_name": "Harsh Patel", "author_url": "", "post_date": "2020-04-09T06:32:36.617000", "content": "Thanks for clarification. Does Keras model.predict() use TPU for batch inference if the model is compiled under TPU strategy?
", "votes": 0, "replies": [] } ] }, { "id": 787867, "author_name": "Hariharan.SJ", "author_url": "", "post_date": "2020-03-27T06:37:42.553000", "content": "Does it matter which image size I use to train the model? the training set has 4 different folders.
", "votes": 0, "replies": [ { "id": 788429, "author_name": "Martin Görner", "author_url": "", "post_date": "2020-03-27T16:58:22.163000", "content": "Yes, usually larger images give a better final accuracy. They are much slower to train on though.\nAlso, when using pre-trained models, these tend to perform best on the image size they have been pre-trained on, even if they can support other sizes.
", "votes": 1, "replies": [] } ] }, { "id": 787277, "author_name": "Jesudas DSouza", "author_url": "", "post_date": "2020-03-26T17:04:59.613000", "content": "Hi, I'm new to Kaggle and just wanted to clarify this statement in the description \"That code must be made public following the end of the competition.\" . Does it mean that I should make the kernel public?
", "votes": 0, "replies": [ { "id": 796506, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-04-03T16:29:41.200000", "content": "@jesudasdsouza To be eligible for the TPU leaderboard prizes, the prospective top 3 winners will need to share the code used to generate their submissions (using machine learning code, fully run on TPUs, including training) publicly. You can wait to do this after the competition closes (so as not to keep your approach private, until the end). To be eligible for the TPU Star prizes, a key aspect we will evaluate on are your upvoted, public contributions, so those notebooks/discussion posts should also be public.
", "votes": 1, "replies": [] } ] }, { "id": 746625, "author_name": "Dmcgow", "author_url": "", "post_date": "2020-02-15T09:30:52.757000", "content": "Amazing
", "votes": 0, "replies": [] }, { "id": 832177, "author_name": "", "author_url": "", "post_date": "2020-05-03T23:39:04.850000", "content": "", "votes": 0, "replies": [] }, { "id": 747532, "author_name": "", "author_url": "", "post_date": "2020-02-16T14:51:56.343000", "content": "", "votes": 0, "replies": [ { "id": 749537, "author_name": "Martin Görner", "author_url": "", "post_date": "2020-02-18T19:01:45.167000", "content": "The training has to happen in the notebook. Using a notebook to read from a file and write the answers into the submission file will not make you eligible for prizes. That said, doing so is still fine. This is a friendly, \"playground\" competition. You are free to experiment with whatever you want to solve the problem at hand. Restrictions are only for the prize slots.
", "votes": 0, "replies": [] }, { "id": 752384, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-02-21T00:47:30.430000", "content": "A correction to Martin's statement. Training does not have to be done in Kaggle notebooks. But per the Prizes page, it must have been trained on TPUs, although that can happen outside of Kaggle. We will be validating that is the case when reviewing the candidates for these awards. Inference and submission does have to be done in Kaggle notebooks.
\n\n> To be eligible for these prizes, your submission must generate predictions from machine learning code fully run on TPUs for both training and inference. That code must be made public following the end of the competition.
", "votes": 0, "replies": [] } ] }, { "id": 829302, "author_name": "Deepak Rajpurohit", "author_url": "", "post_date": "2020-05-01T17:07:02.107000", "content": "thanks
", "votes": 0, "replies": [] }, { "id": 746533, "author_name": "kojingbo", "author_url": "", "post_date": "2020-02-15T06:43:40.713000", "content": "Thank you, is very useful
", "votes": 0, "replies": [] }, { "id": 835872, "author_name": "", "author_url": "", "post_date": "2020-05-06T15:04:29.553000", "content": "thank you
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "744833": "New to machine learning and data science? No question is too basic or too simple. Feel free to start your own thread, or use this thread as a place to post any first-timer clarifying questions for the Kaggle community to help you with! \n\nIf you would consider yourself a beginner but don't know where to get started, let other Kagglers help you take your first steps here!\n\nNew to Kaggle? Take a look at a few videos our very own Dr. Rachael Tatman has put together to learn a bit more about [site etiquette](https://www.youtube.com/watch?v=aIus8si_Et0), [Kaggle lingo](https://www.youtube.com/watch?v=sEJHyuWKd-s), and [how to enter a competition using Kaggle Notebooks](https://www.youtube.com/watch?&v=GJBOMWpLpTQ).\n\nReady to dive into this competition? Review the [Overview Description](https://www.kaggle.com/c/flower-classification-with-tpus/overview) which also provides links to docs and a getting started tutorial.", "745062": ">TPUs read training data exclusively from Google Cloud Storage.\n\nThere is something I want to confirm.\nIs there a charge when using TPU in Notebook?", "832420": "Hi, i just want to mention guide [Submit Prediction](https://www.kaggle.com/c/flower-classification-with-tpus/submit) is a bit confusing for Kaggle newcomers like me.\n\nThe \"Step 1\" says you should click \"Submit\", but i couldn't find it (on Kernel type Notebook). After check all possible button, i found out that i must click \"Save Version\" and choose \"Save & Run All (Commit)\".", "828989": "Like Google Colab,Is there any functionality to connect runtime as TPU in kaggle Notebook?\nBecause After running this below code:\ntry:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver() \n print('Running on TPU ', tpu.master())\nexcept ValueError:\n tpu = None\n\nif tpu:\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n strategy = tf.distribute.get_strategy()\n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync) \n\n\nI got output as below:\n\nREPLICAS:1\n\nAnd After fitting model I got error : \"Your Notebook tried to allocate more memory than its available .It has restarted\"\n\nplease give solution", "824191": "I have a question whether or not the validation set can be used to train the final model in predicting the results?", "815960": "Is it necessary to make modifications to BatchNorm or other layers to run correctly on TPUs? I noticed that the reference Efficientnet models in the [tensorflow/tpu repo](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/utils.py#L89) make some adjustments. If I understood correctly, it looked like the normalization gets calculated on a per replica basis and gets adjusted to calculate across replicas.", "798430": "@juliaelliott @mgornergoogle \n\nCan you guys please confirm that we have to use the **tpu.estimator** to PREDICT as well?", "787867": "Does it matter which image size I use to train the model? the training set has 4 different folders.", "787277": "Hi, I'm new to Kaggle and just wanted to clarify this statement in the description \"That code must be made public following the end of the competition.\" . Does it mean that I should make the kernel public? ", "746625": "Amazing", "832177": "", "747532": "", "829302": "thanks\n", "746533": "Thank you, is very useful", "835872": "thank you\n" } }