nkhuyu

def binary_crossentropy_with_ranking(y_true, y_pred):
    """ Trying to combine ranking loss with numeric precision"""
    # first get the log loss like normal
    logloss = K.mean(K.binary_crossentropy(y_pred, y_true), axis=-1)
    
    # next, build a rank loss
    
    # clip the probabilities to keep stability
    y_pred_clipped = K.clip(y_pred, K.epsilon(), 1-K.epsilon())

nkhuyu

System Design Cheatsheet

Picking the right architecture = Picking the right battles + Managing trade-offs

Basic Steps

1. Clarify and agree on the scope of the system

• User cases (description of sequences of events that, taken together, lead to a system doing something useful)
  • Who is going to use it?
  • How are they going to use it?

nkhuyu

##Perks of Microsoft

####Salary

• Can't go into details, but can't complain at all.

####Health and Wellness Care

• Medical and hospitalization: Two choices for medical insurance. Both with no premium but different deductibles.
  • The two choices on medical plans are a high-deductible health plan (Microsoft puts a chunk of money into the Health Savings Account for you, which covers most of the deductible) and an HMO.
• With the HMO, you pay basically nothing as long as you only go to Group Health doctors. With the high-deductible plan, you're covered under the local Blue Cross provider which means you can go to just about any doctor in the country. For

nkhuyu

What I Wish I'd Known About Equity Before Joining A Unicorn

Disclaimer: This piece is written anonymously. The names of a few particular companies are mentioned, but as common examples only.

This is a short write-up on things that I wish I'd known and considered before joining a private company (aka startup, aka unicorn in some cases). I'm not trying to make the case that you should never join a private company, but the power imbalance between founder and employee is extreme, and that potential candidates would

nkhuyu