80% Inexpensive Travel and 20% Supervised Machine Learning

2018-01-30 @Travel

Empirical evidence suggests that I’m oblivious to most experiences attached to any significant cost. Granted, I’m an amateur even with respect to my own mind and cannot infallibly predict what impact the experience today will have on the future. I can, however, analyze years of past travel experiences along with their features, observing their impact on my memory footprint, and train my internal model to probabilistically predict the value of any new potential experience. (This, by the way, is Supervised Machine Learning.)

From here on, my definition of a valuable experience is one that leaves a lasting impression in the long term. That is, months or even years later I can recall the overall experince highlights or at least the minimal contour, and perhaps reminisce even, with certain nostalgia.

Sometimes I deviate from the model for the sake of acquiring new data. Otherwise, I run the risk of entirely neglecting possible false negatives. First off, any experience erroneously predicted as valuable is a false positive. I would then readjust the model to account for the recently misclassified experience. However, by strictly respecting the classification predicted as valuable, I train a more conservative model, becoming inductively biased towards inflexibility and lack of compromise.

In light of the above, I occasionally entertain a negatively predicted experience, despite my intuition, introducing some stochasticity into the model. This opens the possibility for not only the discovery of a false negative (and the successive model readjustment), but a more liberal, flaw admitting model.

Generally, I encounter false positives with far greater frequency than false negatives. This carries one of two implications. Either

  1. the pool of experiences I find valuable have naturally become more conservative, in which case I need not take special action since the model will gradually eliminate the recurrence of similar false positives, or
  2. I need to increase the model stochasticity and allow a greater percentage of ‘bad decisions’, that is even more aggressively pursue a negatively predicted experience in hope of training a more objective model.

Leaving the data scientific terminology aside, what sorts of travel experiences have I found valuable and lasting, and what sorts have I found artificial and quickly fading?

So how do I continuously travel on an inexpensive basis? First, I spend weeks to months in a place, not wasting financial resources or energy on transit. I pay longer-term rates on lodgings, yet still identify much room for optimization in that regard. Lastly, I simply don’t care for most experiences that cost anything substantial (or often anything at all). In the worst case I can reproduce the value by means of a free alternative. Mostly everything I care for involves daily low-maintenance activities, the people, and resources that naturally surround me.

Questions, comments? Connect.