I am a joyful graduate from the "books era," in which only my teachers were my gurus, no one else, not YouTube or social media. From that perspective, I was a student focused solely on core concepts. With the advent of technology, many more directions, subjects, and concepts have since popped up. Hence, I cannot claim to be an official student of those new subjects; I learned a few concepts on the go, as required.
One of them that is most dear to me is Data Analysis. We talk to each other, you know who?those hidden patterns that data wants to show me and which I wish to retrieve and learn.
Initially, I was only aware of empirical details, which dictated that I first preprocess the data, ensure it is clean and crystal clear, locate impactful attributes, and perform analysis once the data is ready. The focus then was on learning only from historical data. Who else understands the importance of core concepts better than me? But very recently, I understood the importance of recent, fresh data, of course, supported by what happened in the past.
So now, I concretely feel that I must keep performing incremental clustering of ever-growing data, and keep incrementally learning about what data is trying to tell you. This should be done with more focus on current data and trends. I must also shift my focus to impactful rows, data series and time series rather than merely individual attributes.
All these new things will help decision-makers more than just analyzing empirical data. As the data grows now, I must consider the topmost patterns, what exactly they are directing toward. Clubbing the two, analysis generated based on historical data and the freshness added by new add-ons, will prove more fruitful.
On top of that, the individual analyst may use randomization techniques to see what else is possible. I should also analyze worldwide data from the same niche domain to determine what kind of futuristic data may be generated. Thus, instead of one or two, the analyst may have three layers of details to produce optimal results.
These are just thoughts that have popped up. The researcher hidden in me should tap the publications to know if such ensemble or meta-model methods are already proposed at the data level, not the algorithm level.
कोणत्याही टिप्पण्या नाहीत:
टिप्पणी पोस्ट करा