Transfer of ensembles from one account to another.

[Ravi Teja Kankanala]

Hi,

I have recreated the ensembles from one account to another account. I find that the predictions are different.

Please let me know how to fix this.

Thanks,

Ravi.

[Francisco J Martin | BigML]

Ravi,

Each ensemble is different to a previous one in nature as they might use a different sample for each model in the ensemble or even different fields if you used a random decision forest (i.e, "randomize": true). The differences can be accentuated depending on the number of features and the number of instances in your dataset. If you are getting different predictions you may want to increase the number of models in the ensemble.

Anyway, will you please send us the parameters that you use in both cases (number of models, randomize or not, sampling options) and the dates when you create them.

If you can share the ids of both ensembles via sup...@bigml.com. We can check the metadata and see if there's something else going on.

Best, francisco

--
You received this message because you are subscribed to the Google Groups "BigML" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bigml+un...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Ravi Teja Kankanala]

test

[Ravi Teja Kankanala]

HI,

The ensemble which I would like to transfer : https://bigml.com/dashboard/ensemble/5256e5fe035d070aea007477 Date: 10th Oct.

The ensemble which I created with the same base dataset and has different prediction values :  https://bigml.com/dashboard/ensemble/52692c77035d0737bd02db49 Date: 24th Oct

Please let me know the way forward.

Thanks,

Ravi.

On Thursday, October 31, 2013 8:21:54 PM UTC+5:30, Francisco J Martin wrote:

[Francisco J Martin | BigML]

Ravi, 

We'll have a look at the metadata and let you know soon….

In the meanwhile, you want to continue using the previous ensemble to make predictions you can share it via a shared_hash and  then use it locally to run predictions. 

Something like:

curl "https://bigml.io/ensemble/5256e5fe035d070aea007477?$BIGML_AUTH" -X PUT -H "content-type: application/json" -d '{"shared": true}' 

and then 

curl "https://bigml.io/shared/ensemble/<sharedhash>" 

Please feel free to come to https://bigmlinc.campfirenow.com/f20a0 if you need real-time help. 

francisco