rnyak
commented
10 months ago @ralgond NVTabular is easy to use. you can take our existing notebooks and just add your data in it.
if you want to use pandas sure, go for it. but then you need to use schema file yourself manually and correctly, so that you can train a TF4Rec model. Please note that NVTabular generates schema file automatically.
ralgond
How to use Transformers4Rec with pandas
NVTabular is hard to learn, I want to skip it, and load/transform data for Transformers4Rec. Could you please show me how to achive what I want?
For example, now I was developing an news recommendations, which is exactly an session-based recommender, the train data is as following:
user_id | item_id-count | item_id-list | category-list -- | -- | -- | -- 0 | 2 | [2343, 16105] | [26, 281] 1 | 2 | [6678, 29121] | [133, 418] 2 | 2 | [3222, 17493] | [43, 297] 3 | 2 | [3222, 5196] | [43, 99] 4 | 2 | [3789, 4359] | [66, 67] ... | ... | ... | ... 199995 | 7 | [7592, 27484, 28222, 28702, 28515, 33114, 34315] | [136, 409, 412, 412, 412, 437, 442] 199996 | 13 | [1379, 7591, 13521, 13555, 13579, 13515, 13509... | [7, 136, 250, 250, 250, 250, 250, 409, 428, 43... 199997 | 2 | [22678, 22649] | [354, 354] 199998 | 20 | [23642, 24255, 23898, 24254, 24173, 24128, 254... | [375, 375, 375, 375, 375, 375, 389, 389, 395, ... 199999 | 11 | [472, 4245, 5215, 7591, 13521, 16531, 16619, 1... | [4, 67, 99, 136, 250, 281, 281, 281, 281, 297,...200000 rows × 4 columns