orsonadams

GPU Talk from Jane Street: Performance Deep Dive

https://www.youtube.com/watch?v=pHqcHzxx6I8

CPU vs. GPU Performance

• A 2048 x 2048 matrix multiplication takes about 28 ms on 1 CPU core.
• The same operation on 1 GPU takes roughly 209 microseconds ($\mu s$).
  • Question: This is a massive speedup! what specific kind of CPU and GPU were they using for this comparison?

---

orsonadams

# @parsethis
# Basic LRU Cache implementation using a dictionary and a queue
# Goes without saying that you'd not use this, functools.lru_cache is what you want
# TODO, is not threadsafe, not nearly use locking (see functools.lru_cache )

#       hashing keys could be flattened, properly hashed
#       maybe you want a doubly linked list and not a python queue
#       
#       

orsonadams

# List unique values in a DataFrame column
# h/t @makmanalp for the updated syntax!
df['Column Name'].unique()

# Convert Series datatype to numeric (will error if column has non-numeric values)
# h/t @makmanalp
pd.to_numeric(df['Column Name'])

# Convert Series datatype to numeric, changing non-numeric values to NaN
# h/t @makmanalp for the updated syntax!

orsonadams

import numpy as np
from keras.preprocessing.sequence import pad_sequences
from keras.utils import to_categorical
from keras.models import Sequential
from keras.layers import Embedding, Dense, LSTM


def main():

orsonadams

SELECT 
  title, timestamp, score, url
FROM 
  [bigquery-public-data.hacker_news.full] 
WHERE 
  (REGEXP_MATCH(url,r'nytimes.com|cnn.com|hufftingtonpost.com|usatoday.com|reuters.com|politico.com|yahoo.com
                |npr.org|latimes.com|nbcnews.com|cbsnews.com|nypost.com|abcnews.com|breitbart.com|theonion.com
                |newsmax.com|washingtontimes.com|newsweek.com|observer.com|chicago.suntimes.com|theguardian.com
                |buzzfeed.com|wjs.com|abcnews.go.com|news.bbc.co.uk|usatoday.com|bloomberg.com|news.google.com')) 
                AND deleted IS null AND type='story'

orsonadams

orsonadams

orsonadams

from keras.engine.topology import Layer
from keras import initializations
from keras import backend as K

class Attention(Layer):
    '''Attention operation for temporal data.
    # Input shape
        3D tensor with shape: `(samples, steps, features)`.
    # Output shape
        2D tensor with shape: `(samples, features)`.

orsonadams

#!bin/env/python 
# Simple script I used to get make a map from ACS
# survey data variable names to the descriptive table names.

# It made my life a bit easier when searching for table names 
# programmatically

# here is the ACS varibles site:
# http://api.census.gov/data/2014/acs5/variables.html

orsonadams