This is my first computer vision competition but I am overwhelmed right now… I have good amount of computer vision knowledge (theoretical) but less of applied knowledge. Can anyone suggest how to get started with this competition and the concepts I need to know before approaching this competition, so that I can learn a lot from my first competition. Any articles, kernels, concepts to know (list of it) would do…. Any help would be highly appreciated. Thanks…
", "messages": [ { "id": "1116358", "postDate": "12/17/2020 05:25:32", "content": "This is my first computer vision competition but I am overwhelmed right now… I have good amount of computer vision knowledge (theoretical) but less of applied knowledge. Can anyone suggest how to get started with this competition and the concepts I need to know before approaching this competition, so that I can learn a lot from my first competition. Any articles, kernels, concepts to know (list of it) would do…. Any help would be highly appreciated. Thanks…
", "rawMarkdown": "This is my first computer vision competition but I am overwhelmed right now... I have good amount of computer vision knowledge (theoretical) but less of applied knowledge. Can anyone suggest how to get started with this competition and the concepts I need to know before approaching this competition, so that I can learn a lot from my first competition. Any articles, kernels, concepts to know (list of it) would do.... Any help would be highly appreciated. Thanks...", "votes": null }, { "id": "1116392", "postDate": "12/17/2020 06:22:40", "content": "You can start with public kernels, get a baseline score and see yourself in the leaderboard. That would give you a head start and will motivate you to climb further up. You can then try to incorporate your knowledge to improve the kernels you've tried. If stuck, check discussions and there is no other better place to get up to date paper suggestions to implement. Worst case you would learn something new :) So far i learned a lot from others' suggestions and shares. Actively participating in kaggle competition motivates you to actually read, research and implement a lot.
", "rawMarkdown": "You can start with public kernels, get a baseline score and see yourself in the leaderboard. That would give you a head start and will motivate you to climb further up. You can then try to incorporate your knowledge to improve the kernels you've tried. If stuck, check discussions and there is no other better place to get up to date paper suggestions to implement. Worst case you would learn something new :) So far i learned a lot from others' suggestions and shares. Actively participating in kaggle competition motivates you to actually read, research and implement a lot.", "votes": null }, { "id": "1116407", "postDate": "12/17/2020 06:33:34", "content": "Thanks a lot for the suggestions. I would try to do this. Just a quick question, are there any particular concepts I should know for this competition, just suggest me some 2-3 concepts which would give a head start to me :)
", "rawMarkdown": "Thanks a lot for the suggestions. I would try to do this. Just a quick question, are there any particular concepts I should know for this competition, just suggest me some 2-3 concepts which would give a head start to me :)", "votes": null }, { "id": "1116599", "postDate": "12/17/2020 10:01:40", "content": "Get comfortable with tensorflow or pytorch and read some kernels on this competition.
", "rawMarkdown": "Get comfortable with tensorflow or pytorch and read some kernels on this competition.", "votes": null }, { "id": "1116656", "postDate": "12/17/2020 11:22:56", "content": "@atharvaingle Maybe you will find this discussion useful https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/198241
", "rawMarkdown": "atharvaingle Maybe you will find this discussion useful https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/198241", "votes": null }, { "id": "1116760", "postDate": "12/17/2020 12:56:15", "content": "Key concepts would be transfer learning for neural networks, image augmentation and multi-label classification (and eventually things like alternative loss functions, learning rate schedules and different pre-trained architectures). A great (introductory?) course (that is also amazingly free) is fast.ai. They also have a Python package that builds on top of PyTorch (to do really well, one of course has to go beyond the basic options of the library and/or implement completely new things).
\nA nice notebook (it is doing a bunch of relatively advanced things with learning rate schedules and then in the inference notebook test-time-augmentation), but unlike some others has a decent amount of explanations) is this one that does use the fastai library.
", "rawMarkdown": "Key concepts would be transfer learning for neural networks, image augmentation and multi-label classification (and eventually things like alternative loss functions, learning rate schedules and different pre-trained architectures). A great (introductory?) course (that is also amazingly free) is [fast.ai](https://course.fast.ai/). They also have a Python package that builds on top of PyTorch (to do really well, one of course has to go beyond the basic options of the library and/or implement completely new things).\n\nA nice notebook (it is doing a bunch of relatively advanced things with learning rate schedules and then in the inference notebook test-time-augmentation), but unlike some others has a decent amount of explanations) is [this one](https://www.kaggle.com/muellerzr/recreating-abhishek-s-tez-with-fastai) that does use the fastai library.", "votes": null }, { "id": "1116761", "postDate": "12/17/2020 12:56:43", "content": "I was in the same boat not very long ago. I did the MOOCs, I read the books. But I was entirely clueless on how to build a pipeline from where I load the data, do EDA, put it through some model, to create the submission file, and submit to get a score. I was entirely clueless.
\nHow I solved this problem, and what I suggest-
\nI think that will get you started.
", "rawMarkdown": "I was in the same boat not very long ago. I did the MOOCs, I read the books. But I was entirely clueless on how to build a pipeline from where I load the data, do EDA, put it through some model, to create the submission file, and submit to get a score. I was entirely clueless.\n\nHow I solved this problem, and what I suggest-\n* I took an active interest in closed competitions, looked at their data, read their descriptions, and so on.\n* The most important step in ending my mystery was starting *to look at others' Notebooks in those finished competitions*. Since the competitions were finished, you could look at Notebooks which were really high on the Leaderboard. I could study winning solutions, and learn from them.\n* Winners' notebooks will really seem overwhelming at first- nothing like you have seen in the MOOCs. But hold on to it. Copy winners' Notebooks and submit from that if you wish. Then learn by tweaking stuff in that notebook. I suggest that you copy stuff by hand, and try and understand each line. It will take a long time. But bear the pain. It will be worth it.\n* Before doing all these, you have to choose a framework of your choice. In my case, it was PyTorch, and I studied through Notebooks written on PyTorch only. And sometimes the framework was fast.ai.\n* Then dedicate some time to learn the framework of your choice.\n* Actively monitor discussions and publicly available Notebooks related to this competition.\n\nI think that will get you started.", "votes": null }, { "id": "1116766", "postDate": "12/17/2020 13:03:31", "content": "Thank you so much for your suggestions. Will definitely follow them …
", "rawMarkdown": "Thank you so much for your suggestions. Will definitely follow them ...", "votes": null }, { "id": "1116772", "postDate": "12/17/2020 13:07:28", "content": "Thanks for answering… These suggestions are great, will definitely help me :)
", "rawMarkdown": "Thanks for answering... These suggestions are great, will definitely help me :)", "votes": null }, { "id": "1116971", "postDate": "12/17/2020 15:52:04", "content": "Hello @atharvaingle,
\nI recommend you this course for practice fast.ai
", "rawMarkdown": "Hello @atharvaingle,\n\nI recommend you this course for practice [fast.ai](https://www.fast.ai/)", "votes": null }, { "id": "1116983", "postDate": "12/17/2020 16:04:46", "content": "Hi Team, I have background in R and just completed a data science specialisation course from coursera.Iam trying to get into image modelling by trying my hands in the competitions in kaggle.I do not know python .Using Keras and tensorflow in R studio /Kaggle notebook is giving me errors which Iam finding it difficult to debug. ANy guidance on materials I can check out that will be helpful ? Is Python really mandatory if I need to get my hands into Machine learning/Modelling ? Appreciate your inputs.
", "rawMarkdown": "Hi Team, I have background in R and just completed a data science specialisation course from coursera.Iam trying to get into image modelling by trying my hands in the competitions in kaggle.I do not know python .Using Keras and tensorflow in R studio /Kaggle notebook is giving me errors which Iam finding it difficult to debug. ANy guidance on materials I can check out that will be helpful ? Is Python really mandatory if I need to get my hands into Machine learning/Modelling ? Appreciate your inputs.", "votes": null }, { "id": "1117018", "postDate": "12/17/2020 16:40:13", "content": "Are you using TensorFlow for R in RStudio as given in this site or just using the native python TensorFlow. If you are using native Python TensorFlow then no way you can use it in R.
\nAnd regarding your second question that if Python is necessary, I would say yes. It's not that you can't do machine learning and deep learning in R you can surely do it if you want. But Python is widely used and has a great community support. Almost all the frameworks are Python specific, so I would suggest to learn Python. Its very easy to learn and you can pickup the basics in a week. You can even learn it from Kaggle micro courses and there are great tutorials on YouTube for Python. I assume you have some theoretical knowledge of Machine Learning. After getting comfortable with Python, learn the basics of numpy, pandas and matlplotlib. Again these can be learned from Kaggle micro courses itself. Then, pick a dataset and start applying the algorithms. You can do that by sklearn (Python library used mostly for Machine Learning). And if you want to get into Deep Learning, do the fast.ai course as everyone suggested to get a head-start. Then, you can learn theoretical stuffs as you progress further.
\nHope this helps ….
", "rawMarkdown": "Are you using TensorFlow for R in RStudio as given in [this ](https://tensorflow.rstudio.com/) site or just using the native python TensorFlow. If you are using native Python TensorFlow then no way you can use it in R.\n\nAnd regarding your second question that if Python is necessary, I would say yes. It's not that you can't do machine learning and deep learning in R you can surely do it if you want. But Python is widely used and has a great community support. Almost all the frameworks are Python specific, so I would suggest to learn Python. Its very easy to learn and you can pickup the basics in a week. You can even learn it from [Kaggle micro courses](https://www.kaggle.com/learn/overview) and there are great tutorials on YouTube for Python. I assume you have some theoretical knowledge of Machine Learning. After getting comfortable with Python, learn the basics of numpy, pandas and matlplotlib. Again these can be learned from Kaggle micro courses itself. Then, pick a dataset and start applying the algorithms. You can do that by sklearn (Python library used mostly for Machine Learning). And if you want to get into Deep Learning, do the fast.ai course as everyone suggested to get a head-start. Then, you can learn theoretical stuffs as you progress further.\n\nHope this helps ....", "votes": null }, { "id": "1117026", "postDate": "12/17/2020 16:51:50", "content": "Python is not mandatory, but it's simply got the better deep learning ecosystem at the moment. I initially started as an R user exploring Deep Learning using the book on the R keras package (it's a really good book and links to it and other R DL resources can be found e.g. on this website). But I soon realized that I was running into a lot of little problems, could not find other users that had solved similar problems or wanted to use the latest things that were only implemented in some Python repositories. So, for myself I decided to switch 2 years ago.
", "rawMarkdown": "Python is not mandatory, but it's simply got the better deep learning ecosystem at the moment. I initially started as an R user exploring Deep Learning using the book on the R keras package (it's a really good book and links to it and other R DL resources can be found e.g. [on this website](https://blog.rstudio.com/2018/09/12/getting-started-with-deep-learning-in-r/)). But I soon realized that I was running into a lot of little problems, could not find other users that had solved similar problems or wanted to use the latest things that were only implemented in some Python repositories. So, for myself I decided to switch 2 years ago.", "votes": null }, { "id": "1117047", "postDate": "12/17/2020 17:13:18", "content": "Thanks Atharva. I followed two approaches :
\n1) Do the steps in Kaggle R notebook for say Leaf classification competition.Iam encountering errors.
\n2) Installed Keras for R studio in line with the tutorials recommended .I even installed Pyton,Pycharm, packages of Python in my Computer. Tried doing the same image classification say- Fruits classification in my computer and Iam encountering errors.
The problem is google search/ post in forums is not helping me much.
\nThanks much for your inputs on pursuing Python.Iam definitely going to do it.
", "rawMarkdown": "Thanks Atharva. I followed two approaches :\n1) Do the steps in Kaggle R notebook for say Leaf classification competition.Iam encountering errors.\n2) Installed Keras for R studio in line with the tutorials recommended .I even installed Pyton,Pycharm, packages of Python in my Computer. Tried doing the same image classification say- Fruits classification in my computer and Iam encountering errors.\n\nThe problem is google search/ post in forums is not helping me much. \n\nThanks much for your inputs on pursuing Python.Iam definitely going to do it.", "votes": null }, { "id": "1117049", "postDate": "12/17/2020 17:13:57", "content": "Thanks a lot for your inputs.Iam sailing in the same boat now.
", "rawMarkdown": "Thanks a lot for your inputs.Iam sailing in the same boat now.", "votes": null }, { "id": "1117080", "postDate": "12/17/2020 17:55:30", "content": "I am using R, Keras, and Tensorflow for this contest. My results are not yet very good (LB 0.866) but are still improving. I happen to prefer R to Python as my base language, but you may be better off going with Python for the reasons given by the other posters.
\nI have many years of Kaggle experience using R, but this is only my third Code Competition, and it took some effort to get used to the online notebook environment and overcome the various errors and problems I encountered. Currently I train models on one of my own computers that has an Nvidia Quadro GP100 GPU, using the R Keras function 'application_inception_resnet_v2', pretrained with 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5', and adding my own input and output layers as appropriate for this contest. There are probably newer and better networks I could use, and I hope to eventually try some. My input image resolution is currently 384 x 384.
\nI upload multiple models, each trained with a different random number seed, to the cloud as datasets, load those into my notebook, and ensemble their test set predictions together to make submissions.
\nOne book I found quite useful is \"Deep Learning with R\" by Francois Chollet and J. J. Allaire.
\nHope this helps.
", "rawMarkdown": "I am using R, Keras, and Tensorflow for this contest. My results are not yet very good (LB 0.866) but are still improving. I happen to prefer R to Python as my base language, but you may be better off going with Python for the reasons given by the other posters.\n\nI have many years of Kaggle experience using R, but this is only my third Code Competition, and it took some effort to get used to the online notebook environment and overcome the various errors and problems I encountered. Currently I train models on one of my own computers that has an Nvidia Quadro GP100 GPU, using the R Keras function 'application_inception_resnet_v2', pretrained with 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5', and adding my own input and output layers as appropriate for this contest. There are probably newer and better networks I could use, and I hope to eventually try some. My input image resolution is currently 384 x 384.\n\nI upload multiple models, each trained with a different random number seed, to the cloud as datasets, load those into my notebook, and ensemble their test set predictions together to make submissions.\n\nOne book I found quite useful is \"Deep Learning with R\" by Francois Chollet and J. J. Allaire.\n\nHope this helps.", "votes": null }, { "id": "1117089", "postDate": "12/17/2020 18:08:28", "content": "Thanks a lot David.Will check out the Book.
", "rawMarkdown": "Thanks a lot David.Will check out the Book.", "votes": null }, { "id": "1117753", "postDate": "12/18/2020 11:46:57", "content": "You can refer to previous CV kaggle competitions
\nhttps://www.youtube.com/watch?v=1-myowrUhok
\nits an awesome video from Abhishek depicting some of the awesome kaggle competitions of the past.
Thank you.Will check out.
", "rawMarkdown": "Thank you.Will check out.", "votes": null } ], "comments": [ { "id": 1116392, "author_name": "keremt", "author_url": "", "post_date": "12/17/2020 06:22:40", "content": "You can start with public kernels, get a baseline score and see yourself in the leaderboard. That would give you a head start and will motivate you to climb further up. You can then try to incorporate your knowledge to improve the kernels you've tried. If stuck, check discussions and there is no other better place to get up to date paper suggestions to implement. Worst case you would learn something new :) So far i learned a lot from others' suggestions and shares. Actively participating in kaggle competition motivates you to actually read, research and implement a lot.
", "votes": null, "replies": [ { "id": 1116407, "author_name": "atharvaingle", "author_url": "", "post_date": "12/17/2020 06:33:34", "content": "Thanks a lot for the suggestions. I would try to do this. Just a quick question, are there any particular concepts I should know for this competition, just suggest me some 2-3 concepts which would give a head start to me :)
", "votes": null, "replies": [] }, { "id": 1116656, "author_name": "psvishnu", "author_url": "", "post_date": "12/17/2020 11:22:56", "content": "@atharvaingle Maybe you will find this discussion useful https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/198241
", "votes": null, "replies": [] } ] }, { "id": 1116599, "author_name": "zainahmedsharif", "author_url": "", "post_date": "12/17/2020 10:01:40", "content": "Get comfortable with tensorflow or pytorch and read some kernels on this competition.
", "votes": null, "replies": [] }, { "id": 1116760, "author_name": "bjoernholzhauer", "author_url": "", "post_date": "12/17/2020 12:56:15", "content": "Key concepts would be transfer learning for neural networks, image augmentation and multi-label classification (and eventually things like alternative loss functions, learning rate schedules and different pre-trained architectures). A great (introductory?) course (that is also amazingly free) is fast.ai. They also have a Python package that builds on top of PyTorch (to do really well, one of course has to go beyond the basic options of the library and/or implement completely new things).
\nA nice notebook (it is doing a bunch of relatively advanced things with learning rate schedules and then in the inference notebook test-time-augmentation), but unlike some others has a decent amount of explanations) is this one that does use the fastai library.
", "votes": null, "replies": [ { "id": 1116766, "author_name": "atharvaingle", "author_url": "", "post_date": "12/17/2020 13:03:31", "content": "Thank you so much for your suggestions. Will definitely follow them …
", "votes": null, "replies": [] } ] }, { "id": 1116761, "author_name": "truthr", "author_url": "", "post_date": "12/17/2020 12:56:43", "content": "I was in the same boat not very long ago. I did the MOOCs, I read the books. But I was entirely clueless on how to build a pipeline from where I load the data, do EDA, put it through some model, to create the submission file, and submit to get a score. I was entirely clueless.
\nHow I solved this problem, and what I suggest-
\nI think that will get you started.
", "votes": null, "replies": [ { "id": 1116772, "author_name": "atharvaingle", "author_url": "", "post_date": "12/17/2020 13:07:28", "content": "Thanks for answering… These suggestions are great, will definitely help me :)
", "votes": null, "replies": [] } ] }, { "id": 1116971, "author_name": "milobele", "author_url": "", "post_date": "12/17/2020 15:52:04", "content": "Hello @atharvaingle,
\nI recommend you this course for practice fast.ai
", "votes": null, "replies": [] }, { "id": 1116983, "author_name": "gayathrirprog", "author_url": "", "post_date": "12/17/2020 16:04:46", "content": "Hi Team, I have background in R and just completed a data science specialisation course from coursera.Iam trying to get into image modelling by trying my hands in the competitions in kaggle.I do not know python .Using Keras and tensorflow in R studio /Kaggle notebook is giving me errors which Iam finding it difficult to debug. ANy guidance on materials I can check out that will be helpful ? Is Python really mandatory if I need to get my hands into Machine learning/Modelling ? Appreciate your inputs.
", "votes": null, "replies": [ { "id": 1117018, "author_name": "atharvaingle", "author_url": "", "post_date": "12/17/2020 16:40:13", "content": "Are you using TensorFlow for R in RStudio as given in this site or just using the native python TensorFlow. If you are using native Python TensorFlow then no way you can use it in R.
\nAnd regarding your second question that if Python is necessary, I would say yes. It's not that you can't do machine learning and deep learning in R you can surely do it if you want. But Python is widely used and has a great community support. Almost all the frameworks are Python specific, so I would suggest to learn Python. Its very easy to learn and you can pickup the basics in a week. You can even learn it from Kaggle micro courses and there are great tutorials on YouTube for Python. I assume you have some theoretical knowledge of Machine Learning. After getting comfortable with Python, learn the basics of numpy, pandas and matlplotlib. Again these can be learned from Kaggle micro courses itself. Then, pick a dataset and start applying the algorithms. You can do that by sklearn (Python library used mostly for Machine Learning). And if you want to get into Deep Learning, do the fast.ai course as everyone suggested to get a head-start. Then, you can learn theoretical stuffs as you progress further.
\nHope this helps ….
", "votes": null, "replies": [] }, { "id": 1117026, "author_name": "bjoernholzhauer", "author_url": "", "post_date": "12/17/2020 16:51:50", "content": "Python is not mandatory, but it's simply got the better deep learning ecosystem at the moment. I initially started as an R user exploring Deep Learning using the book on the R keras package (it's a really good book and links to it and other R DL resources can be found e.g. on this website). But I soon realized that I was running into a lot of little problems, could not find other users that had solved similar problems or wanted to use the latest things that were only implemented in some Python repositories. So, for myself I decided to switch 2 years ago.
", "votes": null, "replies": [] }, { "id": 1117047, "author_name": "gayathrirprog", "author_url": "", "post_date": "12/17/2020 17:13:18", "content": "Thanks Atharva. I followed two approaches :
\n1) Do the steps in Kaggle R notebook for say Leaf classification competition.Iam encountering errors.
\n2) Installed Keras for R studio in line with the tutorials recommended .I even installed Pyton,Pycharm, packages of Python in my Computer. Tried doing the same image classification say- Fruits classification in my computer and Iam encountering errors.
The problem is google search/ post in forums is not helping me much.
\nThanks much for your inputs on pursuing Python.Iam definitely going to do it.
", "votes": null, "replies": [] }, { "id": 1117049, "author_name": "gayathrirprog", "author_url": "", "post_date": "12/17/2020 17:13:57", "content": "Thanks a lot for your inputs.Iam sailing in the same boat now.
", "votes": null, "replies": [] }, { "id": 1117080, "author_name": "dslate", "author_url": "", "post_date": "12/17/2020 17:55:30", "content": "I am using R, Keras, and Tensorflow for this contest. My results are not yet very good (LB 0.866) but are still improving. I happen to prefer R to Python as my base language, but you may be better off going with Python for the reasons given by the other posters.
\nI have many years of Kaggle experience using R, but this is only my third Code Competition, and it took some effort to get used to the online notebook environment and overcome the various errors and problems I encountered. Currently I train models on one of my own computers that has an Nvidia Quadro GP100 GPU, using the R Keras function 'application_inception_resnet_v2', pretrained with 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5', and adding my own input and output layers as appropriate for this contest. There are probably newer and better networks I could use, and I hope to eventually try some. My input image resolution is currently 384 x 384.
\nI upload multiple models, each trained with a different random number seed, to the cloud as datasets, load those into my notebook, and ensemble their test set predictions together to make submissions.
\nOne book I found quite useful is \"Deep Learning with R\" by Francois Chollet and J. J. Allaire.
\nHope this helps.
", "votes": null, "replies": [] }, { "id": 1117089, "author_name": "gayathrirprog", "author_url": "", "post_date": "12/17/2020 18:08:28", "content": "Thanks a lot David.Will check out the Book.
", "votes": null, "replies": [] } ] }, { "id": 1117753, "author_name": "mrinath", "author_url": "", "post_date": "12/18/2020 11:46:57", "content": "You can refer to previous CV kaggle competitions
\nhttps://www.youtube.com/watch?v=1-myowrUhok
\nits an awesome video from Abhishek depicting some of the awesome kaggle competitions of the past.
Thank you.Will check out.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1116358": "This is my first computer vision competition but I am overwhelmed right now... I have good amount of computer vision knowledge (theoretical) but less of applied knowledge. Can anyone suggest how to get started with this competition and the concepts I need to know before approaching this competition, so that I can learn a lot from my first competition. Any articles, kernels, concepts to know (list of it) would do.... Any help would be highly appreciated. Thanks...", "1116392": "You can start with public kernels, get a baseline score and see yourself in the leaderboard. That would give you a head start and will motivate you to climb further up. You can then try to incorporate your knowledge to improve the kernels you've tried. If stuck, check discussions and there is no other better place to get up to date paper suggestions to implement. Worst case you would learn something new :) So far i learned a lot from others' suggestions and shares. Actively participating in kaggle competition motivates you to actually read, research and implement a lot.", "1116407": "Thanks a lot for the suggestions. I would try to do this. Just a quick question, are there any particular concepts I should know for this competition, just suggest me some 2-3 concepts which would give a head start to me :)", "1116599": "Get comfortable with tensorflow or pytorch and read some kernels on this competition.", "1116656": "atharvaingle Maybe you will find this discussion useful https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/198241", "1116760": "Key concepts would be transfer learning for neural networks, image augmentation and multi-label classification (and eventually things like alternative loss functions, learning rate schedules and different pre-trained architectures). A great (introductory?) course (that is also amazingly free) is [fast.ai](https://course.fast.ai/). They also have a Python package that builds on top of PyTorch (to do really well, one of course has to go beyond the basic options of the library and/or implement completely new things).\n\nA nice notebook (it is doing a bunch of relatively advanced things with learning rate schedules and then in the inference notebook test-time-augmentation), but unlike some others has a decent amount of explanations) is [this one](https://www.kaggle.com/muellerzr/recreating-abhishek-s-tez-with-fastai) that does use the fastai library.", "1116761": "I was in the same boat not very long ago. I did the MOOCs, I read the books. But I was entirely clueless on how to build a pipeline from where I load the data, do EDA, put it through some model, to create the submission file, and submit to get a score. I was entirely clueless.\n\nHow I solved this problem, and what I suggest-\n* I took an active interest in closed competitions, looked at their data, read their descriptions, and so on.\n* The most important step in ending my mystery was starting *to look at others' Notebooks in those finished competitions*. Since the competitions were finished, you could look at Notebooks which were really high on the Leaderboard. I could study winning solutions, and learn from them.\n* Winners' notebooks will really seem overwhelming at first- nothing like you have seen in the MOOCs. But hold on to it. Copy winners' Notebooks and submit from that if you wish. Then learn by tweaking stuff in that notebook. I suggest that you copy stuff by hand, and try and understand each line. It will take a long time. But bear the pain. It will be worth it.\n* Before doing all these, you have to choose a framework of your choice. In my case, it was PyTorch, and I studied through Notebooks written on PyTorch only. And sometimes the framework was fast.ai.\n* Then dedicate some time to learn the framework of your choice.\n* Actively monitor discussions and publicly available Notebooks related to this competition.\n\nI think that will get you started.", "1116766": "Thank you so much for your suggestions. Will definitely follow them ...", "1116772": "Thanks for answering... These suggestions are great, will definitely help me :)", "1116971": "Hello @atharvaingle,\n\nI recommend you this course for practice [fast.ai](https://www.fast.ai/)", "1116983": "Hi Team, I have background in R and just completed a data science specialisation course from coursera.Iam trying to get into image modelling by trying my hands in the competitions in kaggle.I do not know python .Using Keras and tensorflow in R studio /Kaggle notebook is giving me errors which Iam finding it difficult to debug. ANy guidance on materials I can check out that will be helpful ? Is Python really mandatory if I need to get my hands into Machine learning/Modelling ? Appreciate your inputs.", "1117018": "Are you using TensorFlow for R in RStudio as given in [this ](https://tensorflow.rstudio.com/) site or just using the native python TensorFlow. If you are using native Python TensorFlow then no way you can use it in R.\n\nAnd regarding your second question that if Python is necessary, I would say yes. It's not that you can't do machine learning and deep learning in R you can surely do it if you want. But Python is widely used and has a great community support. Almost all the frameworks are Python specific, so I would suggest to learn Python. Its very easy to learn and you can pickup the basics in a week. You can even learn it from [Kaggle micro courses](https://www.kaggle.com/learn/overview) and there are great tutorials on YouTube for Python. I assume you have some theoretical knowledge of Machine Learning. After getting comfortable with Python, learn the basics of numpy, pandas and matlplotlib. Again these can be learned from Kaggle micro courses itself. Then, pick a dataset and start applying the algorithms. You can do that by sklearn (Python library used mostly for Machine Learning). And if you want to get into Deep Learning, do the fast.ai course as everyone suggested to get a head-start. Then, you can learn theoretical stuffs as you progress further.\n\nHope this helps ....", "1117026": "Python is not mandatory, but it's simply got the better deep learning ecosystem at the moment. I initially started as an R user exploring Deep Learning using the book on the R keras package (it's a really good book and links to it and other R DL resources can be found e.g. [on this website](https://blog.rstudio.com/2018/09/12/getting-started-with-deep-learning-in-r/)). But I soon realized that I was running into a lot of little problems, could not find other users that had solved similar problems or wanted to use the latest things that were only implemented in some Python repositories. So, for myself I decided to switch 2 years ago.", "1117047": "Thanks Atharva. I followed two approaches :\n1) Do the steps in Kaggle R notebook for say Leaf classification competition.Iam encountering errors.\n2) Installed Keras for R studio in line with the tutorials recommended .I even installed Pyton,Pycharm, packages of Python in my Computer. Tried doing the same image classification say- Fruits classification in my computer and Iam encountering errors.\n\nThe problem is google search/ post in forums is not helping me much. \n\nThanks much for your inputs on pursuing Python.Iam definitely going to do it.", "1117049": "Thanks a lot for your inputs.Iam sailing in the same boat now.", "1117080": "I am using R, Keras, and Tensorflow for this contest. My results are not yet very good (LB 0.866) but are still improving. I happen to prefer R to Python as my base language, but you may be better off going with Python for the reasons given by the other posters.\n\nI have many years of Kaggle experience using R, but this is only my third Code Competition, and it took some effort to get used to the online notebook environment and overcome the various errors and problems I encountered. Currently I train models on one of my own computers that has an Nvidia Quadro GP100 GPU, using the R Keras function 'application_inception_resnet_v2', pretrained with 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5', and adding my own input and output layers as appropriate for this contest. There are probably newer and better networks I could use, and I hope to eventually try some. My input image resolution is currently 384 x 384.\n\nI upload multiple models, each trained with a different random number seed, to the cloud as datasets, load those into my notebook, and ensemble their test set predictions together to make submissions.\n\nOne book I found quite useful is \"Deep Learning with R\" by Francois Chollet and J. J. Allaire.\n\nHope this helps.", "1117089": "Thanks a lot David.Will check out the Book.", "1117753": "You can refer to previous CV kaggle competitions\nhttps://www.youtube.com/watch?v=1-myowrUhok\nits an awesome video from Abhishek depicting some of the awesome kaggle competitions of the past.", "1118031": "Thank you.Will check out." }, "source": "meta" }