Stata-PyData provides routines for efficiently passing numeric data between Python and Stata. It aims to use interfaces similar to traditional Stata and Python I/O tools to provide the speed of shared memory communication with the transparency of file I/O.

Stata-PyData was designed primarily for use in a project I worked on a Brown in which we found ourselves using Python (in particular the Pandas library built on Numpy) for low level data cleaning and manipulation, and Stata for stastical analysis. We were impressed by the speed and flexibility of Python for handling data manipulation and the ease-of-use of Stata for analysis but were frustrated by the painfully slow process of transferring data between the two using files on the disk. Stata-PyData aims to make that process slightly less painful. While the specific constraints (and lack thereof) of our project at Brown inform many design considerations I hope this code will be generally useful.

Python - Defined in shm.py

shm.write_list(data, dtype, key_seed, info_file='segment_info.txt')

This function writes the list defined in data to a shared memory segment. data must be a list else an exception will be thrown from C. dtype is a string equal to int, float or long which described the data type of data. Important note: lists are expected to be of consistant type. Inconsistently typed lists will result in errors or undefined behavior. key_seed is an integer used in a call to ftok('/tmp', key_seed) to obtain a key for the shared memory segment. info_file is a text file containing information about the shared memory segment needed by other programs to attach and read the segment.

shm.write_frame(frame, info_file='segment_info.txt', key_seed=1)

This is a utility function which calls shm.write_list repeatedly over the columns of a Pandas data frame. Data types are inferred from the first element of each column in the data frame. The value of key_seed is incremented by one each time a new column is written to shared memory.

Stata - Defined in shm_use.ado

shm_use using filename [, clear deallocate compress]

shm_use parses the information contained in filename and reads the corresponding data from shared memory into the Stata data area. The underlying C program is multithreaded using pthreads. A new thread is created for each segment listed in filename.

options              description
-----------------------------------------------------------------------------------
clear                replace data currently in memory
deallocate           deallocate the shared memory segments after import
compress             compress data in memory to the lowest possible storage type

Create data in Python and write it to shared memory using shm.write_frame()

import pandas as pd
import numpy as np
import shm

data = pd.DataFrame(np.random.rand(10000000,3), columns = ['var1', 'var2', 'var3'])
print data.describe()
data.loc[1:10,:]
allocated_segments = shm.write_frame(data, info_file = 'random_integers.txt')

Read the data into Stata

shm_use using random_integers.txt, clear compress deallocate
summarize
list in 1/10