Hi,\nI'm trying to make predictions based on numeric train set with vowpal wabbit, however I dont know how to format lines for vowpal.
\n\nCan I just divide label and features with "|" and features with space, like:\n1 | 0.1 0.2 0.89 ...?
How should string for predictions look?\nCan I just print features divided with space, like:\n0.3 0.42 0.13 0.13 ...?
I read vw docs but it's not clear for me.
", "messages": [ { "id": "133160", "postDate": "08/31/2016 17:59:22", "content": "Hi,\nI'm trying to make predictions based on numeric train set with vowpal wabbit, however I dont know how to format lines for vowpal.
\n\nCan I just divide label and features with "|" and features with space, like:\n1 | 0.1 0.2 0.89 ...?
How should string for predictions look?\nCan I just print features divided with space, like:\n0.3 0.42 0.13 0.13 ...?
I read vw docs but it's not clear for me.
", "rawMarkdown": "Hi,\r\nI'm trying to make predictions based on numeric train set with vowpal wabbit, however I dont know how to format lines for vowpal.\r\n\r\nCan I just divide label and features with \"|\" and features with space, like:\r\n`1 | 0.1 0.2 0.89` ...?\r\n\r\nHow should string for predictions look?\r\nCan I just print features divided with space, like:\r\n`0.3 0.42 0.13 0.13` ...?\r\n\r\nI read vw docs but it's not clear for me.", "votes": null }, { "id": "133175", "postDate": "08/31/2016 19:56:30", "content": "I suggest you try this one:
\n\nhttps://github.com/JohnLangford/vowpal_wabbit/wiki/Input-format
\n\nand then the fastml.com blog:
\n\nhttp://fastml.com/blog/categories/vw/
\n\nMost of the posts in the latter link to code examples in github.
", "rawMarkdown": "I suggest you try this one: \r\n\r\nhttps://github.com/JohnLangford/vowpal_wabbit/wiki/Input-format\r\n\r\nand then the fastml.com blog:\r\n\r\nhttp://fastml.com/blog/categories/vw/\r\n\r\nMost of the posts in the latter link to code examples in github.", "votes": null }, { "id": "133192", "postDate": "08/31/2016 21:31:41", "content": "I read vw tutorial, but there is nothing about prediction arguments.\nSecond link is interesting.\nThank you Konrad! :)
", "rawMarkdown": "I read vw tutorial, but there is nothing about prediction arguments.\r\nSecond link is interesting.\r\nThank you Konrad! :)", "votes": null }, { "id": "133385", "postDate": "09/01/2016 19:46:04", "content": "You need a feature name before the value in libSVM format.
\n\nTo use your example:
\n\n 1 | a:0.1 b:0.2 zoo:0.89\n1 is the target feature, so it doesn't need a name, but the (numeric) input features need it.\nIn this example: there are 3 features: a b zoo\nwith values: 0.1 0.2 0.89 respectively.
\n\nThis answer on SO has the cheat-sheet for feature representation in vw:
\n\n", "rawMarkdown": "You need a feature name before the value in libSVM format.\r\n\r\nTo use your example:\r\n\r\n 1 | a:0.1 b:0.2 zoo:0.89\r\n\r\n1 is the target feature, so it doesn't need a name, but the (numeric) input features need it.\r\nIn this example: there are 3 features: a b zoo\r\nwith values: 0.1 0.2 0.89 respectively.\r\n\r\nThis answer on SO has the cheat-sheet for feature representation in vw:\r\n\r\n https://stackoverflow.com/questions/28640837/vowpal-wabbit-how-to-represent-categorical-features/28643968#28643968", "votes": null }, { "id": "133431", "postDate": "09/02/2016 06:14:38", "content": "A function in the given code might help you to convert csv to VW format\nhttps://github.com/hiransuvrat/machinelearning/blob/master/utilities.py
\n\nNote: This was written couple of years back. You might have to struggle a bit to understand all parameters
", "rawMarkdown": "A function in the given code might help you to convert csv to VW format\r\nhttps://github.com/hiransuvrat/machinelearning/blob/master/utilities.py\r\n\r\nNote: This was written couple of years back. You might have to struggle a bit to understand all parameters", "votes": null }, { "id": "134058", "postDate": "09/04/2016 21:48:41", "content": "There is a utility in vw called "csv2vw" in the /utl/ folder
\n\nRun it like this:
\n\nperl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw -h -- -1 train.csv > trn_vw.txt\nthis would take your "train.csv" file and output a vw format filw as "trn_vw.txt"
\n\n-h means it has a header
\n\n-- means no more arugements becuase you are are about to use:
\n\n-1 means use the last column as your label
\n\nif you run
\n\nperl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw \nA nice helpscreen apears with all the options. That should get you started. a nice hidden gem from the vw team that helps with really big files like these.
", "rawMarkdown": "There is a utility in vw called \"csv2vw\" in the /utl/ folder\r\n\r\nRun it like this:\r\n\r\n perl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw -h -- -1 train.csv > trn_vw.txt\r\n\r\nthis would take your \"train.csv\" file and output a vw format filw as \"trn_vw.txt\"\r\n\r\n-h means it has a header\r\n\r\n-- means no more arugements becuase you are are about to use:\r\n\r\n-1 means use the last column as your label\r\n\r\nif you run\r\n\r\n perl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw \r\n\r\nA nice helpscreen apears with all the options. That should get you started. a nice hidden gem from the vw team that helps with really big files like these.", "votes": null } ], "comments": [ { "id": 133175, "author_name": "konradb", "author_url": "", "post_date": "08/31/2016 19:56:30", "content": "I suggest you try this one:
\n\nhttps://github.com/JohnLangford/vowpal_wabbit/wiki/Input-format
\n\nand then the fastml.com blog:
\n\nhttp://fastml.com/blog/categories/vw/
\n\nMost of the posts in the latter link to code examples in github.
", "votes": null, "replies": [] }, { "id": 133192, "author_name": "tomkowski", "author_url": "", "post_date": "08/31/2016 21:31:41", "content": "I read vw tutorial, but there is nothing about prediction arguments.\nSecond link is interesting.\nThank you Konrad! :)
", "votes": null, "replies": [] }, { "id": 133385, "author_name": "sojourner", "author_url": "", "post_date": "09/01/2016 19:46:04", "content": "You need a feature name before the value in libSVM format.
\n\nTo use your example:
\n\n 1 | a:0.1 b:0.2 zoo:0.89\n1 is the target feature, so it doesn't need a name, but the (numeric) input features need it.\nIn this example: there are 3 features: a b zoo\nwith values: 0.1 0.2 0.89 respectively.
\n\nThis answer on SO has the cheat-sheet for feature representation in vw:
\n\n", "votes": null, "replies": [] }, { "id": 133431, "author_name": "suvrathiran", "author_url": "", "post_date": "09/02/2016 06:14:38", "content": "A function in the given code might help you to convert csv to VW format\nhttps://github.com/hiransuvrat/machinelearning/blob/master/utilities.py
\n\nNote: This was written couple of years back. You might have to struggle a bit to understand all parameters
", "votes": null, "replies": [] }, { "id": 134058, "author_name": "rdslater", "author_url": "", "post_date": "09/04/2016 21:48:41", "content": "There is a utility in vw called "csv2vw" in the /utl/ folder
\n\nRun it like this:
\n\nperl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw -h -- -1 train.csv > trn_vw.txt\nthis would take your "train.csv" file and output a vw format filw as "trn_vw.txt"
\n\n-h means it has a header
\n\n-- means no more arugements becuase you are are about to use:
\n\n-1 means use the last column as your label
\n\nif you run
\n\nperl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw \nA nice helpscreen apears with all the options. That should get you started. a nice hidden gem from the vw team that helps with really big files like these.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "133160": "Hi,\r\nI'm trying to make predictions based on numeric train set with vowpal wabbit, however I dont know how to format lines for vowpal.\r\n\r\nCan I just divide label and features with \"|\" and features with space, like:\r\n`1 | 0.1 0.2 0.89` ...?\r\n\r\nHow should string for predictions look?\r\nCan I just print features divided with space, like:\r\n`0.3 0.42 0.13 0.13` ...?\r\n\r\nI read vw docs but it's not clear for me.", "133175": "I suggest you try this one: \r\n\r\nhttps://github.com/JohnLangford/vowpal_wabbit/wiki/Input-format\r\n\r\nand then the fastml.com blog:\r\n\r\nhttp://fastml.com/blog/categories/vw/\r\n\r\nMost of the posts in the latter link to code examples in github.", "133192": "I read vw tutorial, but there is nothing about prediction arguments.\r\nSecond link is interesting.\r\nThank you Konrad! :)", "133385": "You need a feature name before the value in libSVM format.\r\n\r\nTo use your example:\r\n\r\n 1 | a:0.1 b:0.2 zoo:0.89\r\n\r\n1 is the target feature, so it doesn't need a name, but the (numeric) input features need it.\r\nIn this example: there are 3 features: a b zoo\r\nwith values: 0.1 0.2 0.89 respectively.\r\n\r\nThis answer on SO has the cheat-sheet for feature representation in vw:\r\n\r\n https://stackoverflow.com/questions/28640837/vowpal-wabbit-how-to-represent-categorical-features/28643968#28643968", "133431": "A function in the given code might help you to convert csv to VW format\r\nhttps://github.com/hiransuvrat/machinelearning/blob/master/utilities.py\r\n\r\nNote: This was written couple of years back. You might have to struggle a bit to understand all parameters", "134058": "There is a utility in vw called \"csv2vw\" in the /utl/ folder\r\n\r\nRun it like this:\r\n\r\n perl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw -h -- -1 train.csv > trn_vw.txt\r\n\r\nthis would take your \"train.csv\" file and output a vw format filw as \"trn_vw.txt\"\r\n\r\n-h means it has a header\r\n\r\n-- means no more arugements becuase you are are about to use:\r\n\r\n-1 means use the last column as your label\r\n\r\nif you run\r\n\r\n perl /location_of_vw/vowpal_wabbit-8.1.1/utl/csv2vw \r\n\r\nA nice helpscreen apears with all the options. That should get you started. a nice hidden gem from the vw team that helps with really big files like these." }, "source": "meta" }