* Support more dataframes

[abhijithneilabraham]

-Currently the feature supports csv files only. However, integrating more dataframes is easy. Go through the get_dataframe() method in data_utils.py and include support to detect the incoming file and parse the dataframe from it.

[rhamnett]

It'd be good to just be able to do something like:

agent = Agent(df)

[abhijithneilabraham]

It'd be good to just be able to do something like:

agent = Agent(df)

Sure. We will update it asap.

[rhamnett]

@abhijithneilabraham keep up the great work, looking forward to contributing

[abhijithneilabraham]

Thanks! @rhamnett The repo is updated with the changes you suggested, kindly go through the readme and examples and report any bugs.

[rhamnett]

Hello @abhijithneilabraham! That's great! I gave it a quick test on a sample dataset i created from my own columns and the basic functionality seems to work!

Some support for date types etc - datetime64 etc would be really good and then getting the model to understand what today/yesterday/this week/last week means from the current date.

agent.query_db("how many website visits were there in the last week?")
agent.query_db("how many orders have I had today?")
agent.query_db("how many orders have I had for the week beginning 31st August 2020?")

This kind of thing 👍

[abhijithneilabraham]

Makes sense. In fact, I was already building something like that, but the issue is to map the natural language to datetime without approaching it in a rule based way. Let me think about this and will update asap.

[rhamnett]

Makes sense. In fact, I was already building something like that, but the issue is to map the natural language to datetime without approaching it in a rule based way. Let me think about this and will update asap.

Yes, that's whats going to make it interesting :)

[TheurgicDuke771]

Hi,
What will be the df when use MySQL db (have multiple table)?

[abhijithneilabraham]

Haven't implemeted multi-table support yet. We need to join tables in that case, it seems. Will plan that for future. Thanks 😊 .

[TheurgicDuke771]

thanks for the quick reply. Then df should be select * from table, right?

[abhijithneilabraham]

That's correct. Download a table as a dataframe and pass it to this.

You could also contribute a feature if you can enable something like agent=Agent(db_url, table_name). Refer agent.py,data_utils.py

[TheurgicDuke771]

Sure, will check that. And send you a pr if get anything solid.

[abhijithneilabraham]

Thanks!

[abhijithneilabraham]

Hello @abhijithneilabraham! That's great! I gave it a quick test on a sample dataset i created from my own columns and the basic functionality seems to work!

Some support for date types etc - datetime64 etc would be really good and then getting the model to understand what today/yesterday/this week/last week means from the current date.

agent.query_db("how many website visits were there in the last week?")
agent.query_db("how many orders have I had today?")
agent.query_db("how many orders have I had for the week beginning 31st August 2020?")

This kind of thing 👍

This has been done too! We have a Date class in column_types.py