Python/Pandas how to store forex tick data for analysis

Question

^{I am new to programming, Python and Pandas so hopefully this isn't a silly question.}

I have downloaded some FOREX data from here . A month's worth of data is around 50mil lines in CSV format for all pairs.

I would like to eventually be able to test a strategy across multiple time frames and instruments.

Here is the code I'm using:

file_address = '/Users/Oliver/PyCharm/FX_app/test_data/EURUSD_test.csv'

df = pd.read_csv( file_address,
                  names       = ['Symbol', 'Date_Time', 'Bid', 'Ask'],
                  index_col   = 1,
                  parse_dates = True,
                  converters  = { 'Date_Time': convert_string_to_datetime }
                  )

#                                                               a non-PEP8 format
#                                                               didactic purpose

On anything but a truncated test file this reading in process takes a long time.

• Is there a way I should be storing the data so that Pandas can read the files much faster?
• Is there a limit to the size of the data that Pandas can reasonably handle?

Any help would be greatly appreciated.

Answer 1

I used to play with some cash equity tick-by-tick data (top 30% liquid stocks, over 5 mil records per day). Here is my strategy to deal with file-reading issue using chunksize and hdf5 .

import pandas as pd

# this is your FX file path
file_path = '/home/Jian/Downloads/EURUSD-2015-05.csv'
# read into 10,000 rows per chunk, lazy generator, very fast
file_reader = pd.read_csv(file_path, header=None, names=['Symbol', 'Date_time', 'Bid', 'Ask'], index_col=['Date_time'], parse_dates=['Date_time'], chunksize=10000)

# create your HDF5 at any path you like, with compression level 5 (0-9, 9 is extreme)
Jian_h5 = '/media/Primary Disk/Jian_Python_Data_Storage.h5'
h5_file = pd.HDFStore(Jian_h5, complevel=5, complib='blosc')  

# then write all records into hdf5 file
# this will take a while ... but it emphasizes on re-usability across different IPython sessions 
i = 1
for chunk in file_reader:
    h5_file.append('fx_tick_data', chunk, complevel=5, complib='blosc')
    i += 1
    print('Writing Chunk no.{}'.format(i))

Writing Chunk no.1
Writing Chunk no.2
Writing Chunk no.3
Writing Chunk no.4
...
Writing Chunk no.425


# check your hdf5 file, all 4,237,535 records are there
h5_file

Out[8]: 
<class 'pandas.io.pytables.HDFStore'>
File path: /media/Primary Disk/Jian_Python_Data_Storage.h5
/fx_tick_data            frame_table  (typ->appendable,nrows->4237535,ncols->3,indexers->[index])

# close file IO
h5_file.close()

# the advantage is that after you closing your current session, 
# you can still read the file very quickly when you reopen another session

# reopen your IPython session
Jian_h5 = '/media/Primary Disk/Jian_Python_Data_Storage.h5'
h5_file = pd.HDFStore(Jian_h5)  

%time fx_df = h5_file['fx_tick_data']
CPU times: user 1.93 s, sys: 439 ms, total: 2.37 s
Wall time: 2.37 s

Out[12]: 
                             Symbol     Bid     Ask
Date_time                                          
2015-05-01 00:00:00.017000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:00.079000  EUR/USD  1.1212  1.1212
2015-05-01 00:00:00.210000  EUR/USD  1.1212  1.1213
2015-05-01 00:00:00.891000  EUR/USD  1.1212  1.1213
2015-05-01 00:00:05.179000  EUR/USD  1.1212  1.1213
2015-05-01 00:00:06.257000  EUR/USD  1.1212  1.1213
2015-05-01 00:00:09.195000  EUR/USD  1.1212  1.1213
2015-05-01 00:00:09.242000  EUR/USD  1.1212  1.1212
2015-05-01 00:00:09.257000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:09.311000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:09.538000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:14.177000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:14.238000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:15.886000  EUR/USD  1.1211  1.1212
2015-05-01 00:00:17.122000  EUR/USD  1.1211  1.1212
...                             ...     ...     ...
2015-05-31 23:59:45.054000  EUR/USD  1.0958  1.0959
2015-05-31 23:59:45.063000  EUR/USD  1.0958  1.0958
2015-05-31 23:59:45.065000  EUR/USD  1.0958  1.0958
2015-05-31 23:59:45.073000  EUR/USD  1.0958  1.0958
2015-05-31 23:59:45.076000  EUR/USD  1.0958  1.0958
2015-05-31 23:59:45.210000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:45.308000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:45.806000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:45.809000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:45.909000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:46.316000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:46.527000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:47.711000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:51.721000  EUR/USD  1.0957  1.0958
2015-05-31 23:59:57.063000  EUR/USD  1.0957  1.0958

[4237535 rows x 3 columns]

Not bad, we just need about 2s to read the entire file from HDF5 in future sessions.