Showing posts with label deep learning. Show all posts
Showing posts with label deep learning. Show all posts

Harnessing the Data Lifecycle for Customer Experience Optimization: Streaming Classifications On Twitter Streams

Harnessing the Data Lifecycle for Customer Experience Optimization: Streaming Classifications

For a deeper dive see this past webinar: available here.

In the use case solved for this webinar, I am a Streaming Engineer at an airline, CloudAir.   I need to find, filter and clean Twitter streams then perform sentiment analysis.

Score Models in the Stream to Act

As the Streaming Engineer at CloudAIR I am responsible for ingesting data from thousands of sources, operationalizing machine learning models as part of our streams, running real-time ELT/ETL processes and building event processing systems running from devices, servers and edge nodes. For today’s use case, one of our ML engineers had given me a model that was deployed into one of our production Cloudera Machine Learning (CML) environments. I logged into Cloudera Data Platform (CDP), found the model, tested it, and then extracted the information I need to add this model to our streaming ingest flow for the social media team.

I have been given permissions to access the airline-sentiment workshop in CDP Public Cloud.

I can see all the models deployed in the project I have access to. I see that predict-sentiment is the one I am to use. It is deployed and has 8GB of RAM and 2 vCPU.


I can see that it has been running successfully for a while and I can test it right from the project.

You can see the URL after the POST and the accessKey is in JSON.

Using Cloudera Flow Management (CFM) I am ingesting real-time Twitter streams which I filter for only airline specific data. I then clean and transform these records in a few simple steps.   The next pieces I will need are those two critical values from CML: the Access Key and URL for the model. I will add them to an instance of an ExecuteClouderaML processor.

I am also sending the raw tweet (large JSON files) to a Kafka topic for further processing by other teams.

I also need to store this data to tables for ad-hoc queries. So I quickly spin up a virtual warehouse with Impala for reporting uses. I will put my data into S3 buckets as Parquet files, with an external Impala table on top, for these reports.

Once my environment is ready, which will only take a few minutes, I will launch Hue to create a table.

From the virtual warehouse I can grab the JDBC URL that I will need to add to my Impala Connection pool in CFM for connecting to the warehouse. I will also need the JDBC driver.

From CFM I add a JDBC Controller and copy in the URL, the Impala driver name and a link to that JDBC jar. I will also set my user and password, or Kerberos credentials, to connect.

After having called CML from CFM, I can see the scoring results and now use them to augment my twitter data. The data is added to the attributes for each event and does not affect the current flowfile data.

Now that data is streaming into Impala I can run ad-hoc queries and build charts on my sentiment-enriched, cleaned-up twitter data.

For those of you that love the command line, we can grab a link to the Impala command line tool for the virtual warehouse as well, and query from there. Good for quick checks.

In another section of our flow we are also storing our enriched tweets in a CDP Data Center (CDP-DC) Kudu table for additional analytics that we are running in Hue and in a Jupyter notebook
that we spin up with our CDP-DC CML.

Jupyter notebooks spun up from Cloudera Machine Learning let me explore my data and do some charting, graphs and SQL work in Python3.

One of the amazing features that comes in handy when you have a complex flow that spans a hybrid
environment is to have data management and governance abilities. We can do that with Apache Atlas.
We can navigate and search through Atlas to see how data travels through Apache NiFi, Apache
Kafka, tables and Cloudera Machine Learning model activities like deployment.

Final DataFlow For Scoring

We have a Query Record processor in CFM that analyzes the streaming events and looks for
Negative sentiment by influencers, we then push those events to a Slack channel for our social
media team to handle.

As we have seen, we are sending several different streams of data to Kafka topics for further
processing with Spark Streaming, Flink, NiFi, Java and Kafka Streams applications. Using
Cloudera Streams Messaging Manager we can see all the components of our Kafka cluster
and where our events are as they travel through topics in various brokers. You can see
messages in all of the partitions, you can also build up alerts for any part of your Kafka system.
An important piece is you can trace messages from all of the consumers to all of the producers
and see any lag or latency that occurs in clients.

We can also push to our Operational Database (HBase) and easily scan through the rapidly inserted rows.

This demo was presented in the webinar,
Harnessing the Data Lifecycle for Customer Experience Optimization

Source Code Resources

Queries, Python, Models, Notebooks

Example Cloudera Machine Learning Connector


Predicting Sensor Readings with Time Series Machine Learning

Predicting Sensor Readings with Time Series Machine Learning


Sensor Unit (
  • BME280 temperature, pressure, humidity sensor
  • LTR-559 light and proximity sensor
  • MICS6814 analog gas sensor
  • ADS1015 ADC with spare channel for adding another analog sensor
  • MEMS microphone
  • 0.96-inch, 160 x 80 color LCD
  • Raspberry Pi 4
  • Intel Movidius 2
  • JDK 8
  • MiNIFi Java Agent 0.6.0
  • Python 3

Example Data

{"uuid": "rpi4_uuid_omi_20200417211935", "amplitude100": 0.3, "amplitude500": 0.1, "amplitude1000": 0.1, "lownoise": 0.1, "midnoise": 0.1, "highnoise": 0.1, "amps": 0.3, "ipaddress": "", "host": "rp4", "host_name": "rp4", "macaddress": "dc:a6:32:03:a6:e9", "systemtime": "04/17/2020 17:19:36", "endtime": "1587158376.22", "runtime": "36.47", "starttime": "04/17/2020 17:18:58", "cpu": 0.0, "cpu_temp": "59.0", "diskusage": "46651.6 MB", "memory": 6.3, "id": "20200417211935_7b7ae5da-905b-418b-94f1-270a15dbc1df", "temperature": "38.7", "adjtemp": "29.7", "adjtempf": "65.5", "temperaturef": "81.7", "pressure": 1015.6, "humidity": 6.8, "lux": 1.2, "proximity": 0, "oxidising": 8.3, "reducing": 306.4, "nh3": 129.5, "gasKO": "Oxidising: 8300.63 Ohms\nReducing: 306352.94 Ohms\nNH3: 129542.17 Ohms"}

Easy Deep Learning in Apache NiFi with DJL

Custom Processor for Deep Learning

 Happy Mm.. FLaNK Day!

I have been experimenting with the awesome new Apache 2.0 licensed Java Deep Learning Library, DJL.   In NiFi I was trying to figure out a quick use case and demo.   So I use my Web Camera processor to grab a still shot from my Powerbook webcam and send it to the processor.   The results are sent to slack.

Since it's the holidays I think of my favorite holiday movies:   The Matrix and Blade Runner.   So I thought a Voight-Kampf test would be fun.   Since I don't have a Deep Learning QA piece built yet, let's start by seeing if you look human.  We'll call them 'person'.   I am testing to see if I am a replicant.  Sometimes hard to tell.   Let's see if DJL thinks I am human.


Okay, so it least it thinks I am a person.   The classification of a Christmas tree is vaguely accurate.

It did not identify my giant french bread.

Building A New Custom Processor for Deep Learning

The hardest part of was a good NiFi Integration test.   The DJL team provide some great examples and it's really easy to plug into their models.

ZooModel<BufferedImage, DetectedObjects> model =
                     MxModelZoo.SSD.loadModel(criteria, new ProgressBar())
Predictor<BufferedImage, DetectedObjects> predictor = model.newPredictor()
DetectedObjects detection = predictor.predict(img);

All the source is in github and references the below DJL sites and repos.

Using a New Custom Processor as part of a Real-time Holiday Flow

We first add the DeepLearningProcessor to our canvas.

An example flow:
  • GetWebCameraProcessor:  grab an image from an attached webcamera
  • UpdateAttribute:  Add media type for image
  • DeepLearningProcessor:   Run our DJL deep learning model from a zoo
  • PutSlack:   Put DJL results in a text window in slack
  • PostSlack:  Send our DJL altered image to slack
  • Funnel:   Send all failures to Valhalla

If we example the provenance we can see how long it took to run and some other interesting attributes.

We place the results of our image analysis in attributes while we return a new image that has a bounding box on the found object(s).

 We now a full classification workflow for real-time deep learning analysis on images, could be used for Santa watching, Security, Memes and other important business purposes.

The initial release is available here:
Using library and example code from the Deep Java Library (

Source Code:

And now for something completely different, Christmas Cats: