![\"An](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2020\/01\/datasolutionheader.jpg\")
<\/p>\n<\/p>\n
Welcome to the fifth part of this blog series around using containers for Data Science. In parts one<\/a>,\u00a0two<\/a>,\u00a0three<\/a>, and\u00a0four<\/a>, I covered a number of building blocks that we\u2019ll use here. If this is the first blog you\u2019ve seen, it\u2019s worth skimming the first four parts, or even going back and progressing through them. I make a number of assumptions about your familiarity with Docker, storage, and multi-container applications, covered in these previous blogs.<\/p>\n The\u00a0objective\u00a0of this\u00a0blog is to:<\/p>\n This blog will not always demonstrate good data science practice; I\u2019d rather focus on exposing patterns that are worth being aware of that help provide a catalyst for learning. There are many other sources for performance optimisation, and architectural robustness, but this requires a broader level of understanding than I assume in this article. However, I will usually point out when poor practice is being demonstrated.<\/p>\n For example, some performance patterns that aren\u2019t ideal – the database in the diagram below is a bottleneck and will constrain performance. The remit of this blog is to slowly build on core principles, show how to work with them and use these as a basis for further understanding.<\/p>\n This will be a two-part post. In the first part, we will build the environment locally using docker-compose and make some observations about limitations. In the second, we will migrate the functionality across to Azure Kubernetes Service.<\/p>\n We\u2019ll\u00a0use a simple image classification scenario\u00a0requiring\u00a0a number of\u00a0technical capabilities. These form\u00a0a\u00a0simple\u00a0process that classifies a pipeline of images into one of 10 categories.<\/p>\n The scenario we\u2019ll be building\u00a0assumes\u00a0many typical project constraints:<\/p>\n The overall application is outlined in the diagram above. We want to provide some resilience; For example, a failed categorisation due to a crashed process will be retried. At this stage we won\u2019t provide a highly available database, or an event queue with automatic failover, but we may consider this when we move it to Kubernetes.<\/p>\n <\/p>\n We\u2019ll hold our application under a single\u00a0directory tree. Create the\u00a0containers\u00a0<\/strong>directory, and then\u00a0beneath that, create\u00a0four\u00a0sub directories named\u00a0postgres<\/strong>,\u00a0python<\/strong>,\u00a0rabbitmq\u00a0<\/strong>and\u00a0worker<\/strong>.<\/p>\n <\/p>\n Containers are designed to be disposable processes and stateless. We\u2019ll need to ensure that whenever a container terminates, its state can be remembered. We\u2019ll do that using Docker volumes for persistent storage. As our overall architecture diagram shows, we\u2019ll need this for Postgres, RabbitMQ and to hold our images.<\/p>\n Create\u00a0the\u00a0docker volumes\u00a0and then\u00a0confirm\u00a0they\u2019re there.<\/p>\n <\/p>\n I\u2019ve used a\u00a0publicly available set of images\u00a0for classification –\u00a0the classic\u00a0CIFAR<\/a>\u00a0data set.\u00a0Data sets are\u00a0often\u00a0already post-processed to allow for easy inclusion into machine learning code.\u00a0I found a source\u00a0that has them\u00a0in jpg form, which\u00a0can be downloaded\u00a0here<\/a>.<\/p>\n We\u2019ll first clone the CIFAR image repository, then load those images into a volume\u00a0using\u00a0a tiny alpine container\u00a0and\u00a0show that they have been copied to the persistent volume.\u00a0 We\u2019ll\u00a0also\u00a0use\u00a0this\u00a0volume as part of the process to\u00a0queue and categorise each image.\u00a0 Note that in the text below, you can refer to a running container by the\u00a0prefix of its identity\u00a0if\u00a0it\u00a0is unique.\u00a0 Hence \u2018343\u2019 below refers to the\u00a0container with an ID uniquely beginning with\u00a0\u2018343\u2019.<\/p>\n <\/p>\n We\u2019ll process images by adding them to a queue and letting worker processes simply take them from the queue. This allows us to scale our workers and ensure some resilience around the requests. I\u2019ve chosen RabbitMQ as it\u2019s very easy to use and accessible from many programming languages.<\/p>\n To create the\u00a0RabbitMQ service,\u00a0create a\u00a0Dockerfile\u00a0in the\u00a0containers\/rabbitmq<\/strong>\u00a0directory\u00a0with\u00a0the following:<\/p>\n <\/p>\n Now go into that directory and build it:<\/p>\n <\/p>\n Now start a container based on that image:<\/p>\n <\/p>\n If you\u00a0open up\u00a0a browser and go to localhost:15672, you should see the following:<\/p>\n Log in with\u00a0username\u00a0guest\u00a0<\/strong>and password\u00a0guest<\/strong>, and you\u00a0should\u00a0see something like the following:<\/p>\n This will\u00a0allow us to monitor\u00a0queues.<\/p>\n Go to the\u00a0containers\/python<\/strong>\u00a0directory\u00a0and create a new file called\u00a0fill_queue.py<\/strong>. The code below finds a list of all the images to be categorised and adds it to our queue.<\/p>\n I start at the mounted directory of images, and do a tree walk finding every image (ending in png, jpg, or jpeg). I use the location in the full path to define the expected category is (fNameToCategory), and build up an array of JSON payloads.<\/p>\n I then connect to the Rabbit Server. Note that in this case, that\u00a0HOSTNAME\u00a0<\/strong>is defined as your Docker host\u2019s IP address \u2013 in this case, the Docker host\u2019s \u2018localhost\u2019, because the python container has a different localhost than the RabbitMQ container.<\/p>\n I\u00a0declare a new channel, and queue and publish each\u00a0IMGS entry\u00a0as a separate message.<\/p>\n There is a debugging print to\u00a0show the number of images.\u00a0 If all goes well, you shouldn\u2019t see this\u00a0as it will scroll off the screen. Hopefully, you see\u00a0thousands of messages\u00a0showing progress.<\/p>\n <\/p>\n In the same containers\/python<\/strong> directory, create a Dockerfile for your python engine:<\/p>\n <\/p>\n Now build the Docker image:<\/p>\n <\/p>\n Now, run the container, mounting the volume containing our images and executing our script:<\/p>\n <\/p>\n While this is running, you should see the queued messages increase until it reaches 60,000.<\/p>\n Now click on the \u2018Queues\u2019 link in the RabbitMQ management console, and you will see that those messages\u00a0are\u00a0now\u00a0in the\u00a0‘image_queue<\/strong>‘ queue\u00a0waiting to be requested.<\/p>\n If you now click on the\u00a0image_queue\u00a0link, you\u2019ll get a more detailed view of activity within that queue.<\/p>\n <\/p>\n Now provision\u00a0the\u00a0database environment, which will\u00a0simply\u00a0record categorisation results.<\/p>\n In the\u00a0containers\/postgres<\/strong>\u00a0directory, create a\u00a0Dockerfile\u00a0containing the following:<\/p>\n <\/p>\n In the same directory, create a file called\u00a0pg-setup.sql\u00a0containing the following:<\/p>\n <\/p>\n And build the Postgres container image:<\/p>\n <\/p>\n Start the Postgres service. Note that here we\u2019re mounting a docker volume to hold the persistent data when the container terminates.<\/p>\n <\/p>\n If you now connect to the database, you should see that a table has been created for you. This will contain our categorisation results. Note, the password in this case is \u2018password\u2019 as we specified in the POSTGRES_PASSWORD environment variable when starting the container.<\/p>\n <\/p>\n\n
\n
![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS1.jpg\")
<\/p>\nLet\u2019s\u00a0begin<\/h3>\n
$ mkdir -p containers\/postgres\/ containers\/python\/ containers\/rabbitmq\/ containers\/worker\r\n$ tree containers\r\n\r\ncontainers\/\r\n\u251c\u2500\u2500 postgres\r\n\u251c\u2500\u2500 python\r\n\u251c\u2500\u2500 rabbitmq\r\n\u2514\u2500\u2500 worker<\/pre>\n
Create\u00a0Your\u00a0Persistent\u00a0storage<\/h3>\n
$ docker volume create scalable-app_db_data<\/strong>\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 # for Postgres<\/strong>\r\nscalable-app_db_data\r\n$ docker volume create scalable-app_image_data<\/strong>\u00a0\u00a0\u00a0\u00a0\u00a0 # to hold our images<\/strong>\r\nscalable-app_image_data\r\n$ docker volume create scalable-app_mq_data<\/strong>\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 # for Rabbit data<\/strong>\r\nscalable-app_mq_data\r\n$ docker volume create scalable-app_mq_log<\/strong>\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 # for Rabbit logs<\/strong>\r\nscalable-app_mq_log\r\n$ docker volume ls<\/strong>\r\nDRIVER\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 VOLUME NAME\r\nlocal\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0scalable-app_db_data\r\nlocal\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0scalable-app_image_data\r\nlocal\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0scalable-app_mq_data\r\nlocal\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0scalable-app_mq_log\r\n$<\/pre>\n
Load the\u00a0Source\u00a0Images<\/h3>\n
$ mkdir images\r\n$ cd images\r\n$ git clone https:\/\/github.com\/YoongiKim\/CIFAR-10-images.git\r\nCloning into 'CIFAR-10-images'...\r\nremote: Enumerating objects: 60027, done.\r\nremote: Counting objects: 100% (60027\/60027), done.\r\nremote: Compressing objects: 100% (37\/37), done.\r\nremote: Total 60027 (delta 59990), reused 60024 (delta 59990), pack-reused 0\r\nReceiving objects: 100% (60027\/60027), 19.94 MiB | 2.75 MiB\/s, done.\r\nResolving deltas: 100% (59990\/59990), done.\r\nChecking out files: 100% (60001\/60001), done.\r\n$\r\n$ docker run --rm -itd -v scalable-app_image_data:\/images alpine\r\n343b5e3ad95a272810e51ada368c1c6e070f83df1c974e88a583c17462941337\r\n$\r\n$ docker cp CIFAR-10-images 343:\/images\r\n$ docker exec -it 343 ls -lr \/images\/CIFAR-10-images\/test\/cat | head\r\ntotal 4000\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 954 Dec 22 12:50 0999.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 956 Dec 22 12:50 0998.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 915 Dec 22 12:50 0997.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 902 Dec 22 12:50 0996.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 938 Dec 22 12:50 0995.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 957 Dec 22 12:50 0994.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 981 Dec 22 12:50 0993.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 889 Dec 22 12:50 0992.jpg\r\n-rw-r--r--\u00a0\u00a0\u00a0 1 501\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dialout\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 906 Dec 22 12:50 0991.jpg\r\n$ docker stop 343\r\n343<\/pre>\n
The Queueing Service<\/h3>\n
FROM rabbitmq:3-management\r\n\r\nEXPOSE 5672\r\nEXPOSE 15672<\/pre>\n
$ docker build -t rabbitmq<\/strong> .\r\nSending build context to Docker daemon\u00a0 14.85kB\r\nStep 1\/3 : FROM rabbitmq:3-management\r\n3-management: Pulling from library\/rabbitmq\r\n.\r\n.\r\n.\r\nDigest: sha256:e1ddebdb52d770a6d1f9265543965615c86c23f705f67c44f0cef34e5dc2ba70\r\nStatus: Downloaded newer image for rabbitmq:3-management\r\n---> db695e07d0d7\r\nStep 2\/3 : EXPOSE 5672\r\n---> Running in 44098f35535c\r\nRemoving intermediate container 44098f35535c\r\n---> 7406a95c39b3\r\nStep 3\/3 : EXPOSE 15672\r\n---> Running in 388bcbf65e3f\r\nRemoving intermediate container 388bcbf65e3f\r\n---> db76ef2233d1\r\nSuccessfully built db76ef2233d1\r\nSuccessfully tagged rabbitmq:latest\r\n$<\/strong><\/pre>\n
$ docker run -itd\u00a0 -v \"scalable-app_mq_log:\/var\/log\/rabbitmq\" -v \"scalable-app_mq_data:\/var\/lib\/rabbitmq\" --name \"rabbitmq\" --hostname rabbitmq -p 15672:15672 -p 5672:5672 rabbitmq<\/strong>\r\nf02ae9d41778968ebcd2420fe5cfd281d9b5df84f27bd52bd23e1735db828e18\r\n$<\/pre>\n
![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS2.jpg\")
<\/p>\n![\"The](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS3.jpg\")
<\/p>\n#!\/usr\/bin\/env python\r\nimport pika\r\nimport sys\r\nimport os\r\nimport json\r\n\r\nROOT=\"\/images\"\r\nrLen = len(ROOT)\r\nclasses = ('airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')\r\nHOSTNAME=\"<Enter your Docker host\u2019s IP address>\"\r\n\r\n# Determine the expected category by parsing the directory (after the root path)\r\ndef fnameToCategory(fname):\r\n for c in classes:\r\n if (fname.find(c) > rLen):\r\n return (classes.index(c))\r\n return -1 # This should never happen\r\n\r\nIMGS=[]\r\nfor root, dirs, files in os.walk(ROOT):\r\n for filename in files:\r\n if filename.endswith(('.png', '.jpg', '.jpeg')):\r\n fullpath=os.path.join(root, filename)\r\n cat = fnameToCategory(fullpath)\r\n data = {\r\n \"image\" : fullpath,\r\n \"category\": cat,\r\n \"catName\": classes[cat]\r\n }\r\n message = json.dumps(data)\r\n IMGS.append(message)\r\n\r\nconnection = pika.BlockingConnection(pika.ConnectionParameters(host=HOSTNAME))\r\nchannel = connection.channel()\r\n\r\nchannel.queue_declare(queue='image_queue', durable=True)\r\n\r\nprint(\"Number of Images = \", len(IMGS))\r\n\r\nfor i in IMGS:\r\n channel.basic_publish( exchange='', routing_key='image_queue', body=i,\r\n properties=pika.BasicProperties( delivery_mode=2,\u00a0 )\r\n )\r\n print(\"Queued \", i)\r\n\r\nconnection.close()<\/pre>\nFROM python:3.7-alpine\r\n\r\n# Add core OS requirements\r\nRUN apk update && apk add bash vim\r\n\r\n# Add Python Libraries\r\nRUN pip install pika\r\n\r\nADD\u00a0 fill_queue.py \/<\/pre>\n
$ docker build -t python .\r\nSending build context to Docker daemon\u00a0\u00a0 27.8MB\r\nStep 1\/4 : FROM python:3.7-alpine\r\n---> 459651397c21\r\nStep 2\/4 : RUN apk update && apk add bash vim\r\n---> Running in dc363417cf12\r\n.\r\n.\r\n.\r\nSuccessfully installed pika-1.1.0\r\nRemoving intermediate container b40f1782f0c1\r\n---> 35891fccb860\r\nStep 4\/4 : ADD\u00a0 fill_queue.py \/\r\n---> 17cd19050b21\r\nSuccessfully built 17cd19050b21\r\nSuccessfully tagged python:latest\r\n$<\/pre>\n
$ docker run --rm -v scalable-app_image_data:\/images\u00a0 -it python python \/fill_queue.py\r\n\r\nNumber of Images =\u00a0 60000\r\nQueued\u00a0 {\"image\": \"\/images\/CIFAR-10-images\/test\/dog\/0754.jpg\", \"category\": 5, \"catName\": \"dog\"}\r\nQueued\u00a0 {\"image\": \"\/images\/CIFAR-10-images\/test\/dog\/0985.jpg\", \"category\": 5, \"catName\": \"dog\"}\r\n.\r\n.\r\n.\r\n.<\/pre>\n![\"The](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS4.jpg\")
<\/p>\n![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS5.jpg\")
<\/p>\n![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS6.jpg\")
<\/p>\nProviding a Database Store<\/h3>\n
FROM postgres:11.5\r\n\r\nCOPY pg-setup.sql \/docker-entrypoint-initdb.d\/\r\n\r\nEXPOSE 5432\r\n\r\nCMD [\"postgres\"]<\/pre>\n
CREATE TABLE CATEGORY_RESULTS (\r\n FNAME\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0VARCHAR(1024) NOT NULL,\r\n CATEGORY\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0NUMERIC(2) NOT NULL,\r\n PREDICTION\u00a0\u00a0\u00a0\u00a0NUMERIC(2) NOT NULL,\r\n CONFIDENCE\u00a0\u00a0\u00a0 REAL);<\/pre>\n
$ docker build -t postgres .\r\nSending build context to Docker daemon\u00a0 4.096kB\r\nStep 1\/4 : FROM postgres:11.5\r\n---> 5f1485c70c9a\r\nStep 2\/4 : COPY pg-setup.sql \/docker-entrypoint-initdb.d\/\r\n---> e84511216121\r\n.\r\n.\r\n.\r\nRemoving intermediate container d600e2f45564\r\n---> 128ad35a028b\r\nSuccessfully built 128ad35a028b\r\nSuccessfully tagged postgres:latest\r\n$<\/pre>\n
$ docker run --name postgres --rm -v scalable-app_db_data:\/var\/lib\/postgresql\/data -p 5432:5432 -e POSTGRES_PASSWORD=password -d postgres\r\ndfc9bbffd83de9bca35c54ed0d3f4afd47c0d03f351c87988f827da15385b4e6\r\n$<\/pre>\n
$ psql -h localhost -p 5432 -U postgres<\/strong>\r\nPassword for user postgres:\r\npsql (11.5)\r\nType \"help\" for help.\r\n\r\npostgres=# \\d\r\n List of relations\r\n Schema |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Name\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 | Type\u00a0 |\u00a0 Owner\r\n--------+------------------+-------+----------\r\n public | category_results | table | postgres\r\n(1 row)\r\n\u00a0\r\npostgres=# \\d category_results\r\n Table \"public.category_results\"\r\n Column\u00a0\u00a0 |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Type\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 | Collation | Nullable | Default\r\n------------+-------------------------+-----------+----------+---------\r\n fname\u00a0\u00a0\u00a0\u00a0\u00a0 | character varying(1024) |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 | not null |\r\n category\u00a0\u00a0 | numeric(2,0)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 | not null |\r\n prediction | numeric(2,0)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 | not null |\r\n confidence | real\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 |\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 |<\/pre>\n
The Classification Process<\/h3>\n
![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS7.jpg\")
<\/p>\n![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS8.jpg\")
<\/p>\n![\"A](\"https:\/\/www.microsoft.com\/en-us\/industry\/blog\/wp-content\/uploads\/sites\/22\/2021\/08\/SRappAKS9.jpg\")
<\/p>\n![\"Jon](\"https:\/\/www.microsoft.com\/en-gb\/industry\/blog\/wp-content\/uploads\/sites\/22\/2019\/05\/Jon-Machtynger-150x150.jpg\")
Jon is a Microsoft Cloud Solution Architect specialising in Advanced Analytics & Artificial Intelligence with over 30 years of experience in understanding, translating and delivering leading technology to the market. He currently focuses on a small number of global accounts helping align AI and Machine Learning capabilities with strategic initiatives. He moved to Microsoft from IBM where he was Cloud & Cognitive Technical Leader and an Executive IT Specialist.<\/p>\n