This is my first competition in Kaggle which I came across about 2 weeks ago. My approach was to fork some public kernels and tune the model and hyperparameters. Then I applied various ensemble techniques on the outputs of the tuned models.
\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??
\nThanks in advance.
", "messages": [ { "id": 968461, "postDate": "2020-08-13T03:31:52.067Z", "content": "There are two sides to the question.
\n\n\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??
\n
This is okay on Kaggle, learn a bit reading the kernel and maybe get a medal. Now from a data science perspective you should probably learn to do your own code. A lot of the real life stuff in the field has more to do with being autonomous in going from ideation to production than tuning someone else's work.
", "rawMarkdown": "There are two sides to the question. \n\n> Now, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??\n\nThis is okay on Kaggle, learn a bit reading the kernel and maybe get a medal. Now from a data science perspective you should probably learn to do your own code. A lot of the real life stuff in the field has more to do with being autonomous in going from ideation to production than tuning someone else's work.", "votes": 3, "replies": [ { "id": 968469, "postDate": "2020-08-13T03:45:44.220Z", "content": "@arroqc Thanks for pointing out the two different perspectives and the impact of my approach in real life. I will keep that in mind in the future.
", "rawMarkdown": "@arroqc Thanks for pointing out the two different perspectives and the impact of my approach in real life. I will keep that in mind in the future." } ] }, { "id": 968223, "postDate": "2020-08-12T19:24:25.777Z", "content": "Yo ! I think it is one of the best practices if you are late to a competition. You fork a kernel, setup a baseline, streamline your experiments and if you find something good, share with the community ! 😄💥
", "rawMarkdown": "Yo ! I think it is one of the best practices if you are late to a competition. You fork a kernel, setup a baseline, streamline your experiments and if you find something good, share with the community ! 😄💥", "votes": 1, "replies": [ { "id": 968235, "postDate": "2020-08-12T19:34:00.753Z", "content": "@realsid thanks for your opinion. I think that's exactly public kernels are all about. To share ideas with the community.
", "rawMarkdown": "@realsid thanks for your opinion. I think that's exactly public kernels are all about. To share ideas with the community.", "votes": 1 } ] }, { "id": 968077, "postDate": "2020-08-12T17:25:01.863Z", "content": "i believe it should be fine as long as you are learning something out of it and most importantly giving credit to those people for their work.
", "rawMarkdown": "i believe it should be fine as long as you are learning something out of it and most importantly giving credit to those people for their work. ", "votes": 1, "replies": [ { "id": 968086, "postDate": "2020-08-12T17:33:54.627Z", "rawMarkdown": "", "isDeleted": true }, { "id": 968090, "postDate": "2020-08-12T17:34:34.050Z", "content": "Thanks for your opinion @vinaypratap. I actually did learn some new techniques from these kernels which I am hoping to apply in the future.
", "rawMarkdown": "Thanks for your opinion @vinaypratap. I actually did learn some new techniques from these kernels which I am hoping to apply in the future." } ] }, { "id": 967495, "postDate": "2020-08-12T09:36:57.790Z", "content": "This is my first competition in Kaggle which I came across about 2 weeks ago. My approach was to fork some public kernels and tune the model and hyperparameters. Then I applied various ensemble techniques on the outputs of the tuned models.
\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??
\nThanks in advance.
", "rawMarkdown": "This is my first competition in Kaggle which I came across about 2 weeks ago. My approach was to fork some public kernels and tune the model and hyperparameters. Then I applied various ensemble techniques on the outputs of the tuned models.\n\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??\n \nThanks in advance.", "votes": 2 } ], "comments": [ { "id": 968461, "author_name": "Arnaud Roussel", "author_url": "", "post_date": "2020-08-13T03:31:52.067000", "content": "There are two sides to the question.
\n\n\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??
\n
This is okay on Kaggle, learn a bit reading the kernel and maybe get a medal. Now from a data science perspective you should probably learn to do your own code. A lot of the real life stuff in the field has more to do with being autonomous in going from ideation to production than tuning someone else's work.
", "votes": 3, "replies": [ { "id": 968469, "author_name": "Abu Zahid Bin Aziz", "author_url": "", "post_date": "2020-08-13T03:45:44.220000", "content": "@arroqc Thanks for pointing out the two different perspectives and the impact of my approach in real life. I will keep that in mind in the future.
", "votes": 0, "replies": [] } ] }, { "id": 968223, "author_name": "RealSid", "author_url": "", "post_date": "2020-08-12T19:24:25.777000", "content": "Yo ! I think it is one of the best practices if you are late to a competition. You fork a kernel, setup a baseline, streamline your experiments and if you find something good, share with the community ! 😄💥
", "votes": 1, "replies": [ { "id": 968235, "author_name": "Abu Zahid Bin Aziz", "author_url": "", "post_date": "2020-08-12T19:34:00.753000", "content": "@realsid thanks for your opinion. I think that's exactly public kernels are all about. To share ideas with the community.
", "votes": 1, "replies": [] } ] }, { "id": 968077, "author_name": "Vinay Pratap Singh", "author_url": "", "post_date": "2020-08-12T17:25:01.863000", "content": "i believe it should be fine as long as you are learning something out of it and most importantly giving credit to those people for their work.
", "votes": 1, "replies": [ { "id": 968086, "author_name": "", "author_url": "", "post_date": "2020-08-12T17:33:54.627000", "content": "", "votes": 0, "replies": [] }, { "id": 968090, "author_name": "Abu Zahid Bin Aziz", "author_url": "", "post_date": "2020-08-12T17:34:34.050000", "content": "Thanks for your opinion @vinaypratap. I actually did learn some new techniques from these kernels which I am hoping to apply in the future.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "968461": "There are two sides to the question. \n\n> Now, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??\n\nThis is okay on Kaggle, learn a bit reading the kernel and maybe get a medal. Now from a data science perspective you should probably learn to do your own code. A lot of the real life stuff in the field has more to do with being autonomous in going from ideation to production than tuning someone else's work.", "968223": "Yo ! I think it is one of the best practices if you are late to a competition. You fork a kernel, setup a baseline, streamline your experiments and if you find something good, share with the community ! 😄💥", "968077": "i believe it should be fine as long as you are learning something out of it and most importantly giving credit to those people for their work. ", "967495": "This is my first competition in Kaggle which I came across about 2 weeks ago. My approach was to fork some public kernels and tune the model and hyperparameters. Then I applied various ensemble techniques on the outputs of the tuned models.\n\nNow, I wish to know that, is this bad practice for beginners like me to approach a data science competition by tuning models from public kernels rather than building a model from scratch? If so, can you suggest any alternative approaches??\n \nThanks in advance." } }