These are chat archives for FreeCodeCamp/DataScience

5th
Sep 2018
Jesper Kielsgaard
@jkielsgaard
Sep 05 2018 10:15
hi all. I may have a noob question, but can not google to the answer and think maybe you guys can giv me a more accurate answer. Is the ML using the training data every time to find the result
Ayo Philip
@Frawzey
Sep 05 2018 10:18

Good question:
There are 2 types of learning in ML - supervised & unsupervised.

Supervised learning uses the training data to find the 'results'.

Result = perform its modelling on the test data.
So yes, ML always uses the training data to find the results....in supervised learning.
Jesper Kielsgaard
@jkielsgaard
Sep 05 2018 10:26
Thanks! What about unsupervised is the training data somehow processed and stored in some kind DB?
Ayo Philip
@Frawzey
Sep 05 2018 10:31
Yes & no. Usually there is no training in unsupervised learning because you just want to see clusters and explore the data.
But if you would like to see clusters of that type of data in the future, then you must 'train' it with the old data.
Jesper Kielsgaard
@jkielsgaard
Sep 05 2018 13:17
Thanks for the good answers.
Next question :)
What define machine learning?
I'm thinking of the algorithm/code part here and how it uses its training data.
Is the ML "looking" into its training data and "comparing" it with the input it has received and give a "mathematically" possible output?
Alice Jiang
@becausealice2
Sep 05 2018 17:32
Humans have this incredible skill that has helped us to evolve, "learning", where we are able to analyze input, recognize patterns, and make decisions based on our understanding of those patterns, right?
Machine learning is when computers do that with very little human influence
Also, I want to clarify my stance on Udacity, their Alumni Network that you gain access to after completion of any of their nanodegrees is incredible. I just don't feel that it can make up for how disappointing the course itself was as far as cost is concerned.
Zojja
@zojjak
Sep 05 2018 18:49
its all good. It just seems their costs jumped substantially in the last year or 2 since I last looked
Alice Jiang
@becausealice2
Sep 05 2018 20:21
oh my god
You made me curious so I checked to see how much the nanodegree I did costs now and it's free?
Or they've got some pretty sketchy marketing going on.....
Zojja
@zojjak
Sep 05 2018 20:37
its weird marketing. Most of them cost $1k, some cost $2k
I had enrolled in the data science one a few years ago when it was like $99 per month or similar?
I did the first project but then got sidetracked
They've also partnered with a company to finance so they may say $0 down but you are basically financing the $1k
Ayo Philip
@Frawzey
Sep 05 2018 20:55

Is the ML "looking" into its training data and "comparing" it with the input it has received and give a "mathematically" possible output?

@jkielsgaard Humans take in input mostly in visual & audio formats. For instance, we grow up seeing the alphabets, A-Z and hearing them from our teachers. They train us on how they come together to form words so that we can learn and pronounce words later - even those we weren't taught in class.
Applying the same logic to ML.
ML encompasses the whole process learning from data without being explicitly programmed. The process of showing the data to the algorithm is called training - training the machine to visualize and recognize the data. So that it can also learn with from it and eventually be able to predict/suggest (pronounce) future results (words) without being explicitly taught. In short, computers learn to do things without being repeatedly programmed (instructed).

Eric Leung
@erictleung
Sep 05 2018 23:57
General advice to learn data science: "build things" https://twitter.com/_brohrer_/status/1037058816473550849