As reported by
Kwasi Ohene-Adu
in campfire, there seems to be no way of setting the project. This seems to be a workaround.

Point to the https://github.com/bigmlcom/bigml-java/blob/master/src/main/resources/binding.properties file to explain where to put credentials.

Currently the connection object has a seed property that contains a string used in random procedures like sampling, model configuration etc. Usually, the seeds used in algorithms can be multiple and are not attached to the connection, so I think it's best not to include it as a connection attribute.

Manage operating_points, operating_kinds, etc

• Fix bugs
• Add utility methods: closestInCluster, sortedCentroids
• Add support for optype items

Can we please add them? At the very least, the one for the latest release. Thanks!

Hello,

the lib it's too basic for java usage, it totally miss strong typing, everything returned it's a JSONObject, it shouldn't be like that, that's javascript style nor java.

you should make models of every request/response and return an object for every call, modeled with getter/setter/arraylist/whatever, it shouldn't be the developer to do jsonbobject.get("objects") ... for... etc...

it should really all demanded to libs, with feature like jacksonmapper that convert the json into a pojo

Hi!
When trying to make a prediction on a regression LocalDeepnet, an exception gets thrown at the fillArray() method, specifically a NullPointerException.
This is due to the Deepnet being made from a dataset with almost 2000 fields, the final Deepnet uses only 1000, so when fillArray() tries to iterate over the full list of fields expanding items and finding missings and finds one, it tries to get the field object from the input_fields which doesn't contains it, it retrieves a null and tries to cast it to Number and invoke intVal() on it.

Maybe I'm not yet fully understanding the function objective but I believe fillArray() should only iterate over input_fields as it should find missings on that list, but I'm not sure if the full list of fields is later needed, if that would be the case, then I think there might be an issue with the structure of the model data.
The dataset has no missings and every input is categorical, but the objective field is numeric.
The Deepnet was created through OptiML on the web platform and is being consumed with the bindings.
Thank you in advance! I'm available to provide any more information about my implementation.

java.lang.ClassCastException: java.lang.Long cannot be cast to java.lang.Integer
at org.bigml.binding.LocalCluster.(LocalCluster.java:138)

this.criticalValue = (Integer)cluster.get("critical_value");

There are a couple of documentation problems:

• In the comments of org.bigml.binding.LocalPredictiveModel there's an example that says:

  BigMLClient bigmlClient = new BigMLClient();

it should be BigMLClient.getInstance(), i guess.

• In the docs, the example with the two-string argument's getInstance example is wrong:

   BigMLClient api = BigMLClient.getInstance("foo", "xxxxxxxxxxx");

that does not create a client using a username and an api key, but another thing involving a seed and an storage. The correct invocation would be:

   BigMLClient api = BigMLClient.getInstance("foo", "xxxxxxxxxxx", false);

I actually think having a getInstance taking two strings that are not
username/apikey is very confusing (as demonstrated by the fact that
even our documentation get it wrong) and would change the interface. But
if you don't agree, at the very least the docs should be updated.

(i'm assigning the bug to @leonhwang just because he's the last committer, please reassign as needed)

All the local models use the Serializable interface, however they also extend the ModelFields class which contains a non-static Logger variable. For the local models to be serialisable, the "LOGGER" variable in ModelFields needs to be made static.

Hi,
There's an incompatibility with models and ensembles (I don't know if its the same case with deepnets or any other kind of model), where if the ensemble or any of the models that compose it have more than 1000 fields, the rest is ignored, then when the MultiModel is created and tries to construct every LocalPredictiveModel it finds that there are missing fields and throws an Exception with the message "Some fields are missing to generate a local model. Please, provide a model with the complete list of fields.".

In every ensemble and model theres an property called "fields_meta" that indicates the total amount of fields and the count and offset of the current request, and the API supports indicating these parameters so pagination is used and the full list of fields can be retrieved.
I supose this needs a kind of extensive change as the resources are all downloaded through the same method without checking what is being retrieve, It should check the kind of resource and check for the pagination data "fields_meta" so consecutive requests are done until the complete list of data (fields) is retrieved.

https://bigml.com/api/ensembles?id=filtering-and-paginating-fields-from-an-ensemble

Call Stack for the case of and ensemble with 107 models and 3877 max columns (same as max used fields, altough ensemble.importance size indicates the actual number of used fields, but I ignore if using that number could affect the construction):

java.lang.Exception: Some fields are missing to generate a local model. Please, provide a model with the complete list of fields.

at org.bigml.binding.LocalPredictiveModel.(LocalPredictiveModel.java:129)
at org.bigml.binding.LocalPredictiveModel.(LocalPredictiveModel.java:100)
at org.bigml.binding.MultiModel.(MultiModel.java:92)
at org.bigml.binding.LocalEnsemble.init(LocalEnsemble.java:295)
at org.bigml.binding.LocalEnsemble.(LocalEnsemble.java:154)