Can I use time series data about pytorch-widedeep HOT 6 CLOSED

Hi @LinXin04, yes, it can, you can perhaps take these two notebooks (part 1 and part 2 ) as an example.

In there, we use a sequence of movies watched to predict the next movie a user would watch. But in general, the problem can be frame as to use a sequence of events to predict anything.

Let me know if you have other questions

from pytorch-widedeep.

@jrzaurin
Thank you for your advice. I have read the recommended one, but I feel that it is quite different from my scene.

My scenario is to determine the risk type of the user's current transaction.

The factors that need to be considered include the textual data of the current transaction, tabular data, and the transaction behavior sequence (amount, time) of the previous user in a period of time.

how can i achieve it?

from pytorch-widedeep.

Well, that depends more on you than the library really, in the sense that it depends on how you formulate the problem. At the moment we do not support more than one component per mode, for example, we do not support multiple text components. You seem to have a genuine text component and a second data source that is a sequence, that can also be seen/processed as a text component (since text is...well...a sequence).

Therefore, if you wanted to include these two data sources in the model, unfortunately, this is not something we support now, but we are looking into this since you are not the first to ask (see #125 ) .

Now, if you manage to formulate the problem such that you can represent the sequence also in tabular form and have two data modes (tabular and text) then the library could do it as "usual" (i.e. model = WideDeep(deeptabular = ..., deeptext=...)).

I hope this helps, and let me know if you have more questions :)

from pytorch-widedeep.

@jrzaurin
Thank you. But when i have text and sequence at the same time, i will have two 'deeptext'. but now this is not supported？
for example, now you support:
text_col="review_text"

but not support:
text_cols=["review_text_1", "review_text_2"]

if i have two real texts, then i can concatenate two texts together?

from pytorch-widedeep.

" for example, now you support:
text_col="review_text"

but not support:
text_cols=["review_text_1", "review_text_2"]"

This is correct, we do not support this at the moment, see here: text_col must be a str

If you concatenate texts together and pass then as a single column, then yes, that is 100% fine

We will address this limitation at some point in the near future, I hope...

from pytorch-widedeep.

Thank you very much!

from pytorch-widedeep.