• Tabular data with heterogeneously-typed columns, as in an SQL table or Excel spreadsheet
• Support for many different data types and manipulations including: floating point & integers, boolean, datetime & time delta, categorical & text data.
• Easy handling of missing data
• Powerful groupby, window functions, resampling & aggregations
• Intelligent, data-dependent slicing, indexing and data subsetting
• Many different IO connectors, including: SQL, excel, csv, HDF5, BigQuery, Stata, SAS, JSON

What is Pandas?

• Vectorization for the masses
• ETL
• Fast and Efficient DataFrame
• Interoperability with ecosystem
• Database-like
• User friendly API
• Munging & data prep is a big part of the pipeline

Why Pandas?

• automatic data alignment
• rolling, expanding, and EWM operations
• timeseries ops: fillna, dropna
• resampling & ordered merges
• timezone handling
• date offsets & holiday support
• intelligent interactive indexing

Why Pandas in Finance?

• HDF5
• bcolz
• CSV
• SQL
• JSON
• pickle
• msgpack

I/O & Serialization

http://matthewrocklin.com/blog/work/2015/03/16/Fast-Serialization/

http://odo.readthedocs.org/en/latest/

• dtype segregation
• block memory layout
• computation backends
• cython for critical parts
• hashtable for indexing

What drives pandas?

• algo
• idioms
• built-in / vectorization
  • pandas/numpy
  • bottleneck/numexpr
  • cython
• ad-hoc cython/numba

How to make pandas fast

• apply across the rows
• itertuples/iterrows
• iterative updating

How to make pandas fast

slow

Do's

• have the correct dtypes

• pd.concat

• Categoricals

• Use idioms & builtin

• .apply across columns

Don'ts

• repeated insertions

• micro optimize

• use loops / re-invent the wheel

• .apply across rows

• .applymap

• nest groupby.apply()

• inplace=True

Global-Interpreter Lock

I need even more!

http://continuum.io/blog/pandas-releasing-the-gil

• out-of-core
• parallelism
• DAG semantics

I need even more!

Dask

I need even more!

I need even more!

(((A + 1) * 2) ** 3)

I need even more!

(B - B.mean(axis=0)) + (B.T / B.std())

I need even more!

https://dask.readthedocs.org/en/latest/

http://blaze.pydata.org/blog/2016/02/17/dask-distributed-1/

http://matthewrocklin.com/blog/work/2015/01/06/Towards-OOC-Scheduling