CSCTechTalk

I believe you should start with simple algorithms and get complicated if needed

Blockbuster.. Moving on :oD

@TheyaaMatti Agreed. Maybe simple algorithms won't do the trick, but it's a good place to start.

Even if they do not do the trick, they will give you a good idea how to deep dive into more complicated ones.

So not on their own, visualisation is critical, piles of R code and tables, even with summaries don't grab attention

It can in the right context, but it's not assured. But, wow, it's amazing what can be gleaned from doing relatively simple analyses using data that has never really been analyzed before. There's a wealth of low hanging fruit.

It all depends on the job at hand. You don't need a crane to lift up a car tire for changing, nor will a car-jack move an entire car.

'Context' is the key word. It needs someone who understands and appreciates the softer elements of the market to make it insightful

Agreed. Begin with the business issue, not the algorithm - simple often is sufficient

For me it is less about the simplicity of the algorithm and more about the appropriate one for the task at hand plus significant consideration for ease of use, in both deployment and in interpreting results.

The trick is knowing when you are dealing with a car and when you are dealing with...a skate board?

I believe simple algorithms can lead to Key Executives insight. However, the insights shared need to be presented in the framework of the data used. Easy for executives to make assumptions that may be beyond what the data shows.

@chrismarin Any rules of thumb for how to determine "appropriate"?

Occam's Razor

heh, if only there'd be a simple rule to determine appropriatedness...

@chrismarin Exactly how I approach it. What's the task? What are the data characteristics? Those two questions determine what algorithm(s) is appropriate.

That said, I believe analysis (be it simple or complex) is required to have the best informed upper-level decisions. It may be ignored in favor of other knowledge, but it should still be a part of any competent decision-maker's toolkit.

Gives me insight into the real strength of each team members and based on which I can give them appropriate work.

One of the key factors for choosing an algo is interpretability. A decision tree, where you can can break down each leaf to a set of rules, is going to be less of a black box than Deep Learning.

W/better #BigData search, it could stop me from duplicating work already completed.

one word: efficiency

Assuming you get efficient access and aggregation of the data.

No more in need to chase data scattered in small RDBMS all around the company

a bigger, better picture, hopefully more accurate way to make decisions.

I can share my analysis results without cluttering others' emails.

1. Work with the ideal data instead of the available data. 2. More merging of datasets that are more powerful together than they are apart. 3. Less time spent on data collection. 4. Make my job more enjoyable in general

effective, actionable decision making

Well... this may be a controversial stance, but maybe don't pay your employees in NewSpeak and Empty Promises. That way I wouldn't have to watch the best talent and experience keep leaving CSC. That would have a HUGE impact.

A dashboard to help predict future impacts so pro-active adjustments can be made faster.

Domain semantics (meaning) of the data should be available to the apps. As opposite to most current apps where domain meaning must be interpreted by the end user to make decisions. Consequently, apps could interpret domain rules.

@JenConway1972 To riff on Jen, effective, actionable analysis that I can write about for my reports

That's why I love the ideas, though not necessarily the execution, behind the Semantic Web. All those RDF triples can be joined across disparate data sources and overlaying an OWL ontology means you have a standardized mechanism for categorization.

Pulled better insight from unstructured text

didn't need to be reconciled and cleaned of near matches and duplicates...

Gave me a summary (situational awareness) of what is currently happening across the enterprise.

Gives me insight into the real strength of each team members and based on which I can give them appropriate work.

in the privacy jail cell.

@LoganWilt Amen! Data that was cleaner to begin with!

had been created in a structured and controlled environment, governed by quality and strategic performance parameters

was standardized and comparable

does not change structure frequently, pushing me to change the way I process it

@J_Cordwell Similiar to a comment from Logan. I like this.

had been recorded by systems of control, without human judgement!

If data cleanliness could be fully automated at scale across disparate data sources I would be in heaven.

That is clean, abundant, relevant

...if I had data that was Consistent, Accurate, Adequate, and Timely, my job would be a breeze. Unfortunately, almost none of our data satisfies all 4 criteria at the same time.

Data scientist and meaningful consolidation of the data from various sources

visualization reading -- being able to understand various graphics and charts.

at least decent programming ability in a data-sci-relavent language, basic stats, basic machine learning, visualization skills, curiosity, and at least decent story telling ability

Machine learning, math, multiple industry knowledge

@LoganWilt Being able to interpret statistical results and link to meaningful actions.

Data Science automation will be the table stakes.

@TheyaaMatti Can you elaborate on that?

Yes, #BigData isn't just about the producer. It's also the consumer.

@LoganWilt This is so true. I wish the community attracted more pure consumers who don't have interest in doing data science but still see its value nonetheless

I think we are still waiting on technology that provides an easy interface for the common user, but I agree with @LoganWilt , visualization and understanding the story.

Data Scientists with cross-verticals SME expertise

Apache Hadoop, Apache Spark, SQL, Machine Learning and Data Mining, Statistical and Quantitative Analysis, Data Visualization, Creativity and Problem Solving

Data Architect - To tie it all together.

Machine Learning and Neural Networks. These are by far the most fascinating--and at times, terrifying--technologies.

I think we are in early days still of Big Data, still learning, current tools will obsolete faster than past techs I believe, but having data science skills with vert SME won't if agile and learning.