These are chat archives for FreeCodeCamp/DataScience

19th
Jan 2017
Alice Jiang
@becausealice2
Jan 19 2017 00:05
DoC websites are a chore in every possible way.
They make it as hard for everyone involved as humanly possible, in my experience
Amelia
@apottr
Jan 19 2017 00:08
it was running on a damn IBM mainframe
the multi-page form required a CSRF token in every form leading back from each page
Alice Jiang
@becausealice2
Jan 19 2017 00:20
I've only spent time on them as an "offender search"er so I haven't felt that pain from them :/
// We commit suicide after 3s just to clean up Exploring other peoples' code is always an experience.
Hèlen Grives
@mesmoiron
Jan 19 2017 02:36
@evaristoc about Amazon; I would shift the metric to the auction moment or identify which company over a long period is consistent in quality.. The pipeline of transport and storage must be short and proper. However that said tomatoes can spoil over night even when looking good! Bruises and fruit flies can do serious damage. Milk is perishable but less vulnerable. Unless you can detect spoilage fungas; quality is also kind or non gmo. Food has lots of properties. Some you want from certain countries. Bananas can be classified by their greenness or average brown spots, but brown island spots are warning of bruises and over ripeness. Each good have a different metric set. Ginger is easy and resilient. So ML with decision tree based on visual cues. Picture snap shots taken, some fruits must be checked each day. Mangostein can look good outside while spoiled inside. Well that's for now the flow of ideas. Excuse for the typos it's 3 am or so. Typing in the dark. Good night 💤
Hèlen Grives
@mesmoiron
Jan 19 2017 03:29
@evaristoc the amazon case forgets the difference between selling, subjective preferences and uses. Only in real time I can make a customized decision. Green bananas are horrible in smoothies; nutritional value can vary and spoilage can be smelled sometimes sooner than seen. These way of thinking means that customers taste must change. We get tomatoes in a state unusable for a tasty dish. All the proper ripening etc must be than at home. A human can just match these preferences in real time. Markets have the greatest variety of food states. When buying online the non critical customer suspends their ability to chose properly food states. What we buy is a possible delicious state in the future. Too philosophical?
evaristoc
@evaristoc
Jan 19 2017 10:46

@mesmoiron too philosophycal, indeed :) . But you are close:

ML with decision tree based on visual cues. Picture snap shots taken, some fruits must be checked each day.

If you focus only on strawberries (!):

  • There is a strong correlation between time and the perceived visual quality of the fruit. That visual clue correlates with other sensorial clues. Just let a big sample of strawberries to rotten and take pictures every day of every corner of the fruit.
  • Use all those pictures as your training, validation and test datasets.
  • Decide, together the QA employees, which fruits pass the test and which one not and assign classes accordingly. This is usually a required step and it is referred as "annotation" in text mining.
  • Run an ML over the pictures of fruits. Use likely a convolutional neural network.

If I understood correctly accuracy was about or over 50%. You might still need some QA employees to test and update the datasets as well as double-checking quality, but no much. Amazon can still grow its business without a big burden in preserving quality standards, which is the goal.

Alice Jiang
@becausealice2
Jan 19 2017 20:17
Udacity's just added a deep learning foundations program. They're really getting in on this whole DS/ML craze...
evaristoc
@evaristoc
Jan 19 2017 21:31

@becausealice2 we have a hip! Everyone is up to...

By the way: someone wrote this article in medium freecodecamp:
https://medium.freecodecamp.com/recognizing-traffic-lights-with-deep-learning-23dae23287cc#.zfthw2ro8

Amelia
@apottr
Jan 19 2017 21:44
ty for this article @evaristoc
good stuff
CamperBot
@camperbot
Jan 19 2017 21:44
apottr sends brownie points to @evaristoc :sparkles: :thumbsup: :sparkles:
:cookie: 326 | @evaristoc |http://www.freecodecamp.com/evaristoc