If you see the answers based on the annotations, I quite dont understand its accuracy.
\n\nI used this kernel to load the jsonline file to dataframe, : https://www.kaggle.com/xhlulu/tf-qa-jsonl-to-dataframe
\n\nCode to extract the answers from document_text: \n\nfor i,j in zip(train['question_text'],train.apply(lambda x : x['document_text'][x['long_answer_start']:x['long_answer_end']],1)):\n if j =='':\n continue\n print('Q:'+i,'A:'+j,sep='\\n')\n
I don't understand the answers, they dont make sense to me. Am I missing something here ?
\n\n```\nQ:who is deputy cm of j and k\nA: r of Jammu and Kashmir - Wikipedia
Q:when did the nba 3 second rule start\nA: ve - second rule of food hygiene , or th
\n\nQ:where does the phrase bob's your uncle come from\nA: H1> Bob 's your uncle
Jump to : navigation , search For other uses , se\n\nQ:what year did the movie deuces come out\nA: Directed by Jamal Hill
\n\nQ:who played the bank robber in dirty harry\nA: ited States <
\n\nQ:where is the best farming soil in the us and where did all the dirt come from\nA: Black Dirt Region is located in southern Orange County , New
\n\nQ:who sings with jackson brown on the load out\nA: nning on Empty A-side `` Stay '' Released D
\n\n```
", "messages": [ { "id": "668420", "postDate": "11/08/2019 12:14:59", "content": "If you see the answers based on the annotations, I quite dont understand its accuracy.
\n\nI used this kernel to load the jsonline file to dataframe, : https://www.kaggle.com/xhlulu/tf-qa-jsonl-to-dataframe
\n\nCode to extract the answers from document_text: \n\nfor i,j in zip(train['question_text'],train.apply(lambda x : x['document_text'][x['long_answer_start']:x['long_answer_end']],1)):\n if j =='':\n continue\n print('Q:'+i,'A:'+j,sep='\\n')\n
I don't understand the answers, they dont make sense to me. Am I missing something here ?
\n\n```\nQ:who is deputy cm of j and k\nA: r of Jammu and Kashmir - Wikipedia
Q:when did the nba 3 second rule start\nA: ve - second rule of food hygiene , or th
\n\nQ:where does the phrase bob's your uncle come from\nA: H1> Bob 's your uncle
Jump to : navigation , search For other uses , se\n\nQ:what year did the movie deuces come out\nA: Directed by Jamal Hill
\n\nQ:who played the bank robber in dirty harry\nA: ited States <
\n\nQ:where is the best farming soil in the us and where did all the dirt come from\nA: Black Dirt Region is located in southern Orange County , New
\n\nQ:who sings with jackson brown on the load out\nA: nning on Empty A-side `` Stay '' Released D
\n\n```
", "rawMarkdown": "If you see the answers based on the annotations, I quite dont understand its accuracy.\n\nI used this kernel to load the jsonline file to dataframe, : https://www.kaggle.com/xhlulu/tf-qa-jsonl-to-dataframe\n\nCode to extract the answers from document_text: \n```\nfor i,j in zip(train['question_text'],train.apply(lambda x : x['document_text'][x['long_answer_start']:x['long_answer_end']],1)):\n if j =='':\n continue\n print('Q:'+i,'A:'+j,sep='\\n')\n```\n\nI don't understand the answers, they dont make sense to me. Am I missing something here ?\n \n```\nQ:who is deputy cm of j and k\nA: r of Jammu and Kashmir - WikipediaI'm not sure what's going on for you, but this is my approach:
\n\ndocument_text.split(' ')[answer_start:answer_end]
I get these:\nwho had the most wins in the nfl ['tom', 'brady']\nhow i.met your mother who is the mother ['tracy', 'mcconnell']
", "rawMarkdown": "I'm not sure what's going on for you, but this is my approach:\n\n`document_text.split(' ')[answer_start:answer_end]`\n\nI get these:\nwho had the most wins in the nfl ['tom', 'brady']\nhow i.met your mother who is the mother ['tracy', 'mcconnell']", "votes": null }, { "id": "668889", "postDate": "11/09/2019 02:56:08", "content": "Sorry, I thought the annotation was character level 😅 .
", "rawMarkdown": "Sorry, I thought the annotation was character level 😅 .", "votes": null }, { "id": "680526", "postDate": "11/24/2019 20:18:53", "content": "Why are some long_answer_start - long_answer_end larger than the number of total number of tokens?? Is the data correct?
", "rawMarkdown": "Why are some long_answer_start - long_answer_end larger than the number of total number of tokens?? Is the data correct?", "votes": null } ], "comments": [ { "id": 668652, "author_name": "afederation", "author_url": "", "post_date": "11/08/2019 17:11:19", "content": "I'm not sure what's going on for you, but this is my approach:
\n\ndocument_text.split(' ')[answer_start:answer_end]
I get these:\nwho had the most wins in the nfl ['tom', 'brady']\nhow i.met your mother who is the mother ['tracy', 'mcconnell']
", "votes": null, "replies": [ { "id": 668889, "author_name": "messdupfreak", "author_url": "", "post_date": "11/09/2019 02:56:08", "content": "Sorry, I thought the annotation was character level 😅 .
", "votes": null, "replies": [] } ] }, { "id": 680526, "author_name": "nectario", "author_url": "", "post_date": "11/24/2019 20:18:53", "content": "Why are some long_answer_start - long_answer_end larger than the number of total number of tokens?? Is the data correct?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "668420": "If you see the answers based on the annotations, I quite dont understand its accuracy.\n\nI used this kernel to load the jsonline file to dataframe, : https://www.kaggle.com/xhlulu/tf-qa-jsonl-to-dataframe\n\nCode to extract the answers from document_text: \n```\nfor i,j in zip(train['question_text'],train.apply(lambda x : x['document_text'][x['long_answer_start']:x['long_answer_end']],1)):\n if j =='':\n continue\n print('Q:'+i,'A:'+j,sep='\\n')\n```\n\nI don't understand the answers, they dont make sense to me. Am I missing something here ?\n \n```\nQ:who is deputy cm of j and k\nA: r of Jammu and Kashmir - Wikipedia