So you have 100,000 rows of data to analyze, you cannot afford to hire a Data Scientist and you do not have the required skills.Have no fear, IBM Watson Analytics have come to the rescue.
Watson Analytics is IBM's attempt to replicate what Apple has done in computing to data science. They have managed to make the subject simple enough for mere mortals (with little or no knowledge of Data Science) like you and me to understand.
The software has sections that it uses to work on data just as shown in the image above, they are namely:
Explore: This section deals with data exploration, which is just having a feel of data in an attempt to find out what the data is all about. The section has charts that will help provide insight as to what the data is all about. It's an array of charts options that will paint a good picture of the data. This section allows users to probe data for insights that will inform decision. Once the data is uploaded and refined, Watson dissects data and comes up questions that users might be interested in finding answers to.
Clicking on the question that best matches objectives, will reveal detailed answers with charts. The software will produce lots of graphs pertaining to the question asked. Clicking on a graph reveals a more detailed information. The image below shows graphs made from the data we tested with.
Predict: This section is basically linear regression in graphic user interface. It users predict values using variables in the datasets. Dependent variables are predicted using independent variables. Watson examines the quality of the dataset and scores it. This enables users to make changes that ensures accurate predictions. The software then explains that datasets using all sorts of graphs and tables showing things like skewness of data, Outliers, Box graphs etc. For users that are novices in linear regression (which is most of us), Watson is able to point out variables that predict other variables. In my dataset, Item variable drives Unit cost, as shown below.
It also shows field associations , which is another way of showing which variables are correlated just as shown below.
It also shows degree and accuracy of predictability in a circular diagram just like the one shown below. It also offers alternatives where necessary, like in my dataset Rep provides a better prediction for item, with predictive strength of 99.6%
With these kind of information, users now know what variables to tweak to get results from other variables. Will discuss other sections in subsequent blogs.
Enjoy Watson.
No comments:
Post a Comment