nifi-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andy Christianson <>
Subject Re: MiNiFi C++ and Tensorflow future plans?
Date Thu, 05 Apr 2018 15:08:10 GMT
Hash: SHA1


> I was wondering what is the roadmap and future plans of MiNiFi C++ agent
> regarding Tensorflow processors. With the three that are mentioned in the
> article it is possible to classify images on edge but I would like to know if
> other processors from TF will be included so we can train a neural network on
> edge.
> Also, if you share with me a list of features that will be implemented in the
> near futures, it would be very helpful.

Glad to hear you are interested in the project. The current plan is to cover
the common edge inference use-cases. This includes general ML inference tasks
as well as computer vision tasks including object classification, object
detection (multiple output FlowFiles with bounding boxes + class),
classification/anomaly detection in log files, time series anomaly detection
(think temperature sensors).

The benefit of using MiNiFi - C++ rather than just a pure TensorFlow model is
that all of the usual NiFi techniques like routing on attribute (i.e. object
class or other ML-inferred metadata), sending to cloud endpoints (S2S to a NiFi
instance, or ingest into a Kafka or MQTT queue, etc.), and arbitrary scriptable
actions (ExecuteScript processor) are all fairly simple to do.

MiNiFi - C++ is a community-driven Apache project, so it ultimately will
include any FlowFile -> tensor or tensor -> FlowFile processor that is
developed to satisfy community requirements.

What use cases do you have in mind? MiNiFi - C++ is nearing its 1.0 release and
as such most common use cases, including deep neural networks or even training
on the edge (given sufficient resources, i.e. GPU) should be possible without
too much custom programming effort. We'd like to get your input/feedback as a
potential community participant so that MiNiFi will become more useful to
everyone over time.


Andy I.C.
Version: GnuPG v2.0.22 (GNU/Linux)


​Sent from ProtonMail, Swiss-based encrypted email.​

View raw message