Train and deploy ML models at scale using Azure Machine Learning | INT134C

WE ARE SO GLAD TO HAVE YOU JOIN ON TRAINING AND DEPLOYING ML MODELS FOR THIS SESSION, WE HAVE A DEDICATED WE REQUEST YOU TAKE ADVANTAGE OF I’LL BRING UP A BUNCH OF QUESTIONS THE Q AND A SESSION AT THE END, CARTACIO. THE FLOOR IS ALL YOURS I’M FROM WASHINGTON. >> MY NAME WASHINGTON, SUPER EXCITED FOR THE AGENDA, WE ARE GOING TO GIVE YOU AND HOW TO DEPLOY MODELS, FOCUS WE’LL SHOW YOU HOW TO USE NATURAL BERT-BASED MODEL TO FINE TUNE THIS PROBLEMS, TO PREDICT AND FORECAST NEED TO BE ABLE TO ANSWER THESE DEPLOY THAT MODEL AS WELL, THEN AND COMMIT QUESTIONS THROUGHOUT AS SOON AS WE ARE DONE WITH OUR WITH A WIDE VARIETY OF COMPANIES HELP PEOPLE ACHIEVE A DIGITAL TRANSFORMATION, A DIGITAL TRANSFORMATION ACROSS PREDICTIVE MAINTENANCE, FRAUD DETECTION, MORE. AZURE MACHINE LEARNING PROVIDES HELP ALL PEOPLE OF SKILL LEVELS UPLOAD YOUR DATA, SPECIFIC WHICH OR WHICH TASK TO PERFORM, IT WILL PICK THE BEST ONE. YOU CAN EASILY IF YOU WOULD LIKE A LITTLE BIT MORE DESIGNER TO DRAG AND DROP DIFFERENT OWN CODE. THEN SIMPLY DEPLOY THAT WE HAVE HOSTED JUPITER NOTEBOOKS OPEN SOURCE MACHINE LEARNING LIBRARY, ANY ARBITRARILY COMPLEX PYTHON CODE CHILD AS WELL. HAVE MLOPS SOLUTION THE DEVOPS CONCEPTS FOR YOUR MACHINE OF WHAT TYPE OF SCENARIO YOU’RE FOR, THE PROCESS IS ALWAYS MORE BY PREPARING YOUR DATA AND THEN AND THIS TAKES SOME AMOUNT OF GUESS OF DIFFERENT SOLUTIONS TO HELP ACCELERATE BEST MACHINE MODEL, ONCE YOU GET TEST YOUR CODE BEFORE YOU DEPLOY THE MACHINE LEARNING MODELS INTO CAPABILITIES TO AND DEPLOY TO THE HOW WELL THEY ARE PERFORMING AND PROCESS END TO END USING OUR MACHINE AHEAD AND DIVE IN TO SEE EXACTLY HERE WE HAVE THE MACHINE LEARNING NEEDS TO GET PRODUCTIVE RIGHT AWAY WHICH IS KIND OF GUESS AND CHECK WE ARE TRYING TO ACHIEVE A SCENARIO QUESTIONS, AND I HAVE TRIED THIS THIS IS A PARTICULAR RUN OF THIS CLUSTER I RAN THIS ON. AND EVEN THIS MACHINE LEARNING MODEL. HAVING THE INPUT DATASET IS SUPER HELPFUL

WHAT DATA WAS USED TO TRAIN THE REGULATORY REASONS, SOMETIMES YOU DATASET YOU USED. YOU CAN ALSO SEE LOGGED DURING THIS EXPERIMENT RUN AND ONCE I’VE CROWNED THE MODEL, IN THE MODEL REGISTRY, DIFFERENT LINK BACK TO THE EXPERIMENT RUN AND I HAVE THE FULL LINEAGE, TRACEABILITY, AS WELL. SO WE CAN SEE THE SCORING MORE. LET’S GO AHEAD AND GO INTO WE CAN SEE, I CAN USE JUPITER NOTEBOOKS WE’LL CLASSIFY AND ROUTE STACK OVERFLOW MODEL. FIRST WE’LL CONNECT TO THE YOU CAN THINK OF EVERYTHING YOU MODELS AND MORE. WE CAN ACCESS THAT THE JUPITER NOTEBOOK OR ANYWHERE WE’LL GRAB A HANDLE TO THE DATASET STACK STACK OVERFLOW QUESTIONS SO USE THE BUILT-IN CAPABILITIES HERE YOU MIGHT USE, I CAN QUICKLY OPEN SO I’LL TOGGLE OVER HERE, AND WE’LL IN THE NATIVE JUPITER EXPERIENCE FOR THE STACK OVERFLOW QUESTIONS DATASET. I’LL GRAB THE HANDLE TO TO USE THIS TO SUBMIT THIS JOB TO CLUSTER. WHEN I SUBMIT THIS JOB, I’M DONE, SO I CAN PAY AS LITTLE LEARNING MODEL. TO LOWER MY COST VM’S, IT WILL USE SPOT PRICING FOR MY COSTS. TO GET STARTED WITH TRAINING EXPERIMENT, IT’S LIKE A CONTAINER ITERATIONS OR RUNS OF THIS EXPERIMENT TOGETHER TO SEE WHICH ONE IS PERFORMING IN ANY ARBITRARILY COMPLEX PYTHON MODEL, IN THIS CASE, I’M USING A TO DISTRIBUTE THE MA SLEEP LEARNING I’LL INPUT MY DATASET AS WELL AND THEN I CAN SEE DIRECTLY HERE IN WELL THE MACHINE LEARNING MODEL ALL MY LOGS AS WELL. IF I WOULD I’VE WALKED AWAY FROM MY MACHINE, AS I SHOWED YOU EARLIER. OR I CAN PEOPLE ON MY TEAM SO THEY CAN MONITOR OF MACHINE LEARNING MODELS BENEFIT DIFFERENT INPUT LEARNING VALUES, AND CHECK TYPE OF APPROACH. I DON’T TO INPUT. AZURE MALONE LEARNING ITERATING OVER DIFFERENT COMBINATIONS BEST MACHINE LEARNING MODEL, YOU OPTIMIZE FOR AND WHEN YOU WANT TO BEST ACCURACY POSSIBLE IN THIS PARTICULAR SUITE, AWE’LL SEE AGAIN I CAN MONITOR IT ACTUALLY RAN MULTIPLE JOBS TO THOSE DIFFERENT INPUT VALUES, AND FAIRLY POORLY, AS YOU MENTIONED, GET DIFFERENT RESULTS. IN THIS CASE TO THE OTHERS, THAT IT ACTUALLY COMPUTE COSTS THAT WEREN’T GOING THE BEST RUN, THEN WE CAN GO AHEAD

DIRECTLY IN LINE HERE, AND SO THIS FOR EXAMPLE. THIS WILL ENABLE YOU LATER AND DEPLOY THEM TO THE CLOUD LEARNING MODEL, FIRST WE’LL WRITE COMPLEX BUSINESS LOGIC YOU WOULD TO INCLUDE AN FUNCTION AND A RUN DATA AND THEN YOU CAN USE BUILT-IN DEPLOY THIS PAGE PAGE TO A KUBERNETES WORK SPACE. USING THE COMBINATION — PROVIDED, AS WELL AS A CONFIGURE THE DOCKER CONTAINER AND STORE THAT ONCE WE HAVE DEPLOYED THAT DOCKER THE REST API, RI, THAT WILL ENABLE WRITE A SIMPLE FUNCTION THAT TAKES THE RUST API. WE CAN SEE I HAVE OVERFLOW QUESTIONS, WE CAN THEN THAT THE MODEL LOOKED LIKE IT PERFORMED ACCURACY AND BASED ON INPUT VALUES, WITH A HIGH PROBABILITY. SOMETIMES IN DEPLOYING THIS MODEL AND USING TO GET THE BEST QUALITY RESULTS TO MONITOR IT AND MAKE SURE THE AND WE’LL USE MACHINE LEARNING PIPELINES AS NEW INPUT DATA COMES INTO THE CHANGE AND ADD NEW SERVICES OR WE IN DIFFERENT WAYS, WE’LL CONTINUOUSLY MACHINE LEARNING PIPELINES ALLOW COMPUTE IN THE PROCESS, IN THIS PREPARE THE DATA THAT’S INCOMING WE CAN SEE I HAVE A REFERENCE TO THE SAME HYPER DRIVE CONFIGURE STEPS SCRIPTS TO PREPARE THE DATA, AND RESULTS AND REGISTER THE MODEL IN INTO A PIPELINE AND THEN I’LL PUBLISH API I INVOKED TO RUN THIS PIPELINE EXTERNAL AGENT THAT I WANT TO CALL THAT OR I CAN USE THE BUILT-IN SCHEDULING THE MACHINE LEARNING PIPELINE, THEN I CAN MONITOR HOW WELL THAT’S PERFORMING STEPS OF THIS MACHINE LEARNING PIPELINE IS COMPLETED AND THE NEW VERSION YOU CAN SEE WE HAVE A COUPLE OF LEARNING MODELS. NOW, BEFORE WE AUTOMATED MACHINE LEARNING TO FORECAST WE HAVE ANY QUESTIONS FROM THE AUDIENCE AREA FOR QUESTIONS, SO KEEP YOUR FOR YOU, CHRIS, CAN ANYONE IN THE COMPUTE AS THEY WANT AND HOW CAN A GREAT QUESTION. SO AZURE MACHINE IT ADMINISTRATIVE TYPE SCENARIOS EXPERIENCES, YOU CAN USE ROLE-BASED PEOPLE TO CREATE THE COMPUTE AND CAN GET CONCURRENTLY, SO YOU CAN CAN USE THAT FUNCTIONALITY TO GRANT TO USE THE COMPUTE, BUT NOT NECESSARILY CREATE NEW CLUSTERS OR MAKE THE TO MAKE THEM. IN ADDITION TO THE

SOME OF THE KEY WAYS YOU CAN MANAGE GO TO THE AUTOMATED MACHINE LEARNING SO AS WE TALK ABOUT THE MA HALLOWEEN BEING TIME CONSUMING AND QUITE COMPLEX ATTRIBUTE TO THAT. THE FIRST IS ABOUT THAT, WE ARE TALKING ABOUT SOMING VALUES IN YOUR DATA OR ADD TO ENRICH TRAINING DATA. WE HAVE TUNING WHERE THE SEARCH IS EXTREMELY APPLY TO A PROBLEM AND MANY DIFFERENT THAT WILL APPLY. SO HOW DO WE SIMPLIFY GOING TO PROPOSE, WE AUTOMATE MACHINE LET’S TAKE A QUICK LOOK AT HOW THIS WITH USER INPUTS, COME IN WITH YOUR AND CONSTRAINTS, THIS COULD BE HOW LIKE THREE HOURS, ONCE YOU HAVE AS YOU LIKE, IT WILL HANDLE THE HANDLING THE MISSING VALUES LIKE INFORMATION TO YOUR DATA SUCH AS ONCE THIS IS DONE AND YOUR DATA ALGORITHM SELECTION. IT WILL INTELLIGENTLY AND HOW ARE ROUE HYPER PARAMETER THE BEST MODEL BUT ALL THE MODELS ALL OF THESE MODELS, YOU’RE GOING WHEN WE TALK ABOUT ANALYSIS, WE MOST POPULAR MODEL IS AN ABILITY THIS TELLS YOU WHAT FEATURES IN RESULTS THE MOST. WE ARE TRYING QUESTIONS, THE THING THAT MIGHT PERHAPS THAT TEAM HAS A REALLY LARGE SEE A LOT LESS. YES WEB BROWSER PROCESS AND A LITTLE BIT ABOUT EXPLANATIONS, TO LEARN HOW THIS WORKS LIVE. HERE I’M ACTUALLY IN THE AUTOMATED AUTHORING START A NEW AUTOMATED ML RUN. HERE IMPORTANT PART OF THE MACHINE LEARNING THREE COLUMNS HERE, THE FIRST IS THE NEXT IS A TAG AND THE LAST IS I HAVE MULTIPLE TEAMS AND HAVE THEIR WE HAVE QUICKLY LOOKED AT THE DATA GET STARTED. HERE I’M GOING TO CONFIGURE TO HOUSE MY RUN, I CAN SELECT AN HERE YOU’LL CREATE A NEW EXPERIMENT COLUMN, THIS IS PURELY WHAT I’M THE VOLUME. FINALLY, I SELECT THE ABOUT THIS, YOU CAN CREATE A NEW TIME TO FOCUS ON THE TASK. OUR TASK WE WILL LET AUTOMATED KNOW WHICH NEXT WE HAVE THE OPTIONAL GROUP A SECOND, WE CONDENSED ALL OF OUR MEANS ALL THE DIFFERENT INFORMATION WE NEED TO LET AUTOMATED ML KNOW IT’S ONLY WORKING WITH ONE FILE WE WANT TO FORECAST PER TEAM. NEXT THIS IS HOW FAR INTO THE FUTURE CAN DO THIS NOW OR YOU CAN ENTER AUTOMATIC DEEP LEARNING, YOU’LL ON THE PREPARATION STATE. PRE-PROCESSING WHILE THIS GETS STARTED, YOU’LL TAB. CLICK ON A PRE-RUN EXPERIMENT DATASET. HERE I CAN SEE A BEST MODEL SEE MY BEST MODEL, ET CETERA, I THIS CAN TELL ME MORE ABOUT HOW

MY DATA AND WHETHER OR NOT ANY FREQUENCY TAB, WE CAN SEE ALL OF THE DIFFERENT DOWNLOAD OR EXPLAIN OUR BEST MODEL CAN CLICK DEPLUS, HERE WE HAVE TO CALL THIS DEPLOYMENT, THEN WE CHOOSE CONTAINER INSTANCE, THAT’S ALL YOU AND DOE EMPLOY, WE REFER TO THIS EVERYTHING IS PREPACKAGED FOR YOU, SCORING SCRIPT OR FILE. I CAN ACTUALLY DEPLOYMENTS, YOU CAN SEE I HAVE AND THE ONE I JUST DID. NOW THAT TO GET PREDICTIONS, I’LL POP OVER HERE IN THIS DATA FLOW, I HAVE DATE, YOU ONLY KEPT IT HERE TO KIND OF FOR A LITTLE BIT OF ANALYSIS, THESE I SAID I CAUSE CONFIGURING MY RUN, CLICK ON AI INCITE . I CAN SEE THE POWER BI INTELLIGENTLY MAPS THE THE MODEL EXPECTS, DATA AND TEAM SO NOW POWER BI IS QUERYING THE AND GOING TO BRING BACK THE PREDICTION OF — ANALYZE SOME OF THIS, WE CAN THE MODEL PREDICTED IS ACTUALLY TRAIN THIS MODEL FOR ALL THAT LONG POWER BI, WE CAN GENERATE REPORTS, SERIES REPORT SHOWING THE DIFFERENT PLACE TO SHOW INFORMATION THAT YOU WE HAVE DONE EVERYTHING FROM SEEING ASPECTS OF THE BUILDING A MACHINE THAT WITH AUTOMATED MACHINE LEARNING, HOW TO PRESENT THAT TO AN AUDIENCE SABINA, GREAT SHOWING HOW TO USE INTO THE Q AND A, I WANTED TO SHOW CONDEMAND SOURCES AT THE LINK BEFORE STARTED AS WELL. WE HAVE LEARNING GET CERTIFIED IN HOW TO USE MACHINE OUT, DO WE HAVE ANY QUESTIONS FROM ARE NOW HEADING TO THE Q AND A SESSION, KNOW WHEN TO TRY REGULAR ML SUCH A GREAT QUESTION, I’LL TAKE THAT OF — A MORE MANUAL CODE PERSPECTIVE AS A SMALL PART IN THE MAJOR WORK ML, YOU CAN AUTOMATIC THE SELECTED AUTOMATED MA SLEEP LEARNING TO GENERATE USE PYTHON SCRIPTS AND OTHER RESOURCES MODEL. NOW IF YOU’RE COMING FROM WHERE PERHAPS BUILDING MODELES IS MA SHEARN LEARNING ALLOWS YOU TO TO BUILD THE MODELS, THE AUTOMATED >> KEEP YOUR QUESTIONS COMING. THE WE USE TENSOR FLOW WITH AZURE? >> YOU CAN USE TENSOR FLOW IN AZURE, MACHINE LEARNING WE ARE OPEN AND LEARNING INSIDE. INCLUDING TENSOR SABINA, BETWEEN HYPER PARAMETER WHAT’S THE DIFFERENCE THERE? >> THE PROS OF AUTOMATED MACHINE LEARNING, THESE PROCESSESES. IF YOU WANT THAT YOU WANT TRUE FLEXIBILITY TO DEFINE EARLY, WHAT SCORES ARE YOU USING,

SOMETHING LIKE HYPER DRIVE, IT’S THE CODE NEEDED TO GET THE INFORMATION VERSUS WITH AUTOMATED MACHINE LEARNING, THE BEST PATH. I SUGGEST TRYING >> COOL. WHAT ARE THE MAIN REASONS TO TRAIN MY ML MODEL INSTEAD OF POWERFUL, TOO? >> YEAH. I ACTUALLY MYSELF. AZURE MACHINE LEARNING FULLY OR DESK TOP TO TRAIN MACHINE LEARNING LOCALLY. THE PYTHON SDK CAN CONNECT LOGGING THE KEY METRICS AND KEEPING AND THEN — LOG FILES, FOR EXAMPLE YOUR NETWORK ARCHITECTURE FOR DNN’S, UP ON LARGER DATASET AND SCALE OUT GPU’S OR DISTRIBUTED CPU CLUSTERS DATA BEATS BETTER ALGORITHMS. SO BE IMMENSELY HELPFUL. WHEN YOU’RE EASIER TO COLLABORATE WITH OTHERS, WHAT THE REST OF YOUR TEAM IS WORKING AS WELL. >> REALLY COOL, OH, I LOVE OF THE BEST MODEL, THE ML. >> THAT’S TWO ANSWERS TO THIS, THE FIRST IS ANY OF THE MODELS THAT AUTOMATED TO DEVELOP THAT MODEL. HOWEVER, OF BUILDING WHAT WE REFER TO — IN TRAINING. SO IF YOU COME THROUGH YOU CAN CLICK AND WE WILL GENERATE OR FURTHER WORK ON THIS MODEL. >> ABOUT THE DEMO CONTENT, CAN WE GET OUT ON GITHUB TODAY. HOPEFULLY CAN FOR AZURE MACHINE LEARNING, GITHUB, BUNCH OF SAMPLES, THOSE WE WALKED FOR SABINA, YOU MENTIONED FORECASTING FORWARD? >> THAT’S A REALLY GREAT A LITTLE BIT TO YOUR SCENARIO, USUALLY THE — SO YOU HAVE ABOUT FOUR YEARS PREDICT TWO YEARS, THIS DEPENDS SO IF YOU KNOW WEATHER IS PART OF PREDICT OUT AS FAR AS YOU KNOW THE CLARITY. >> COOL. DOES IT SUPPORT THE MODEL? >> YEAH. USING THE MACHINE THE PROCESS TO RUN TO TRAIN, VALIDATE CADENCE BASED ON TIME FOR THE MACHINE CHANGES TO THE UNDER-ALLOWING DATA DATA, IT CAN RERUN THAT PIPELINE DATA AND SO YOU, AGAIN, THERE’S TO RUN IF THE DATA HASN’T CHANGED, POP LINE, VALIDATE WHETHER THE NEW AND AUTOMATICALLY ROLL IT OUT USING IS THE AUTOMATION OF THE PROCESSES QUESTION, IS IT WORTH LEARNING PYTHON? IF YOU ARE A . NET DEVELOPER, FOR LIKE ML. NET THAT ENABLE YOU TO AS WELL. RIGHT NOW, THE BROADEST SCIENCE, THE WIDEST VARIETY OF BUILT-IN REALLY EASY TO DO DATA SCIENCE, PYTHON IS BECOMING SUPER POPULAR, WOULD ENCOURAGE LEARNING IT AS WELL THE SESSION, BACK TO YOU, C. S IT’S BEEN A PLEASURE TO PRESENT OF BUILD

OUT THE PEAKS OF INDIVIDUAL USE,

OUT ALL STATION OF THIS SAME ENERGY SCENE A MASSIVE INCREASE OF COMPUTE MASSIVE. SO THAT’S A GOOD THING YEAH. JUST BRINGING IT BACK, WHEN THINK OF IT AS DIRTY, YOU THINK IT AS CLEAN, WE PLUG SOMETHING INTO OUT, NO DUST, NO CHALK, NO NOTHING, DON’T GET DIRTY, BUT IT’S ONE OF ONE OF THE STATS I USE, 49 PERCENT WORLD ARE FROM THE PRODUCTION OF STATISTIC, IT SLAPPED ME IN THE AS WELL BECAUSE, YOU KNOW, YOU THINK THE THINGS I CAN DO WITH ELECTRICITY, I THINK — ONE OF THE THINGS, THE IF YOU’RE STUCK WHAT TO DO, CHOOSE AND THEN ALWAYS CHOOSE THE OPTION AS WELL, IF YOU CAN BE MORE ENERGY — EFFICIENT WITH ELECTRICITY, YOU THE CARBON BACK TO THE TOPIC AGAIN, AND I — FASCINATING TOPIC, ETHINK INTENSITY. I THINK THAT’S WHY WE IN THE PRINCIPLE, CONSUME ELECTRICITY WHAT IS CARBON INTENSITY. >> CARBON WE MENTIONED BEFORE, YOU CAN GET SOURCES AND SOME FORMS ARE GOING MUCH MORE CARBON EMITTED FOR THE BURNING LOTS OF COAL, THAT HAS MUCH PANELSES AND WIND TURBINES, YOU BEARING IN MIND, IN MANY CASES WHEN MIX, SO, YEAH, YOU CAN SAY I WANT MANY WAYS, IF YOU DO NOT HAVE ACCESS THEIR THERELY THE CARBON INTENSITY ON WHERE YOU ARE IN THE WORLD AND SO IN A GOOD EXAMPLE MIGHT JUST DIFFERENT CARBON INTENSITY BECAUSE SOLAR, FOR EXAMPLE, YOU’LL HAVE OF BALANCE THE GRID TO BASICALLY POWER TO — THAT WE CAN USE IS IN THAT WE HAVE, IN MANY CASES, YOU’LL ON THE GRID, YOU’LL END UP EVERYTHING OFTEN SEE HIGH PRICED, HIGH CARBON INTO THE GRID. IF YOU CAN FIND WAYS THEN YOU OFTEN WILL PAY LESS, BUT SUCH AN EXTENT, THERE ARE CERTAIN NEGATIVE ENERGY PRICES. LARGELY ENERGY BEING GENERATED AND THE PEOPLE TO FIND WAYS TO GET PEOPLE TO HELP USED TO BALANCING, TO KIND OF CREATE WE CAN TAKE ADVANTAGE OF AS DEVELOPERS, GREENER PLACES OR RUN AT DIFFERENT SERVICES TO RUN IN THE LONG RUN WORK. I’LL HAND IT BACK BEFORE I QUITE EXCITING AND THERE’S A LOAD NOW. I’LL HAND IT OVER TO YOU. >> ONE OF THE TOPICS THAT FASCINATES THE — THE — THOSE ARE NUMBER BASICALLY THE GRAMS PER KILOWATT HOUR, THIS IS, AS YOU SAID, IN DIFFERENT REGIONS, MORE COAL OR GAS OR SOMETHING ELSE, IN OTHER REGIONS, IT MAY BE LESS, SIDE OF THE THING, AS THE DEMAND

AND SOLAR INCREASE, YOU CAN GET INTENSITY AND JUST WITHOUT CHANGE, CHANGING ANY LINE OF CODE, NOTHING, WHEN OR WHERE YOU RUN YOUR WORKLOAD, THAT EMIT LESS CARBON. THAT’S WHY CARBON INTENSITY AS WELL, THAT GIVES MENTION SOME SPECIFIC EXAMPLES? YOU WORK WITH — THERE ARE SOME THERE’S A GROUP WHO RELEASED A PAPER WHAT THEY DID, THEY BASICALLY BUILT KNEW WHEN ELECTRICITY WAS GREEN — I THINK THEY — I THINK THIS CLOUDS AND BASICALLY USED THIS INFORMATION DEPENDING ON WHERE ENERGY WAS CHEAP THIS INTO THE — INTO KUBERNETES RUNNING, AND THIS HAS — YOU NOW THERE’S A COMPANY CALLED HELLO KIND OF ALLOW USE TO TAKE PART IN LIKE, SAY, KUBERNETES OR DIFFERENT FOUND WAYS TO ALLOW YOU TO SELL IT WERE, IN THE SAME WAY PEOPLE FOR GENERATING POWER ANDEL IT THE COMPANY AND BUILDING THIS ON A LOAD IN A NUMBER OF LARGE COMPANIES NOW, BECOMING MORE ACCESSIBLE, WHEN YOU THINGS WORK IN THINGS LIKE CLOUD, HOW WE PRICE AND USE ENERGY AROUND HERE, LIKE IN EUROPE, FOR EXAMPLE, TOUGH GREEN GRID, WHEN THERE IS THEY CAN SELL IT TO ANOTHER COUNTRY’S HAVE THE INTERNET COMPRISED WITH THERE’S WORK TO DO SOMETHING LIKE OF INTERNET OF POWER, AS IT WERE ANY KIND OF UNDERSTANDING OF INTERNET, SO MUCH INTERESTING STUFF HERE, IS THE THING — IT’S BASICALLY EVERYTHING REALLY FASCINATING. IT’S AMAZING THE REAL WORLD OF THIS INCREDIBLE INTERESTING QUESTIONS COMING THROUGH COULD ASK YOU SOME OF THIS. ONE CONTRIBUTE TO THIS, THIS BEING, ENGINEERING AND BECOME PART OF SUSTAINABLE IS — FEW POINTERS, THERE IS ACTUALLY TALKING ABOUT THIS A LOT. THERE’S DESIGN INTEREST GROUP. THERE’S ALSO, YOU CAN SEE A SIGN. THERE IS ALSO YOU CAN LEARN ABOUT THIS NOW ACTUALLY DESIGN FOR SUSTAINABILITY, CAME BOOK COMING OUT BY THE AUTHOR — WEBSITES, HIS COMPANY. GENERALLY, HAVE — WE SPEND A BUNCH OF TOM, THIS FIELD, BECAUSE YOU ARE SEEING — >> HI, WELCOME TO BUILD. THANK