saving figures

Hello,

    Is it possible to 'save' a matplotlib figure object using something like the python pickle module? Basically, I'd like to save the 'figure' as a file so that I can open it and manipulate it if something is wrong. Alternative suggestions are welcome.

Thank you,

--Jordan Atlas

Jordan Atlas wrote:

Hello,

    Is it possible to 'save' a matplotlib figure object using something like the python pickle module? Basically, I'd like to save the 'figure' as a file so that I can open it and manipulate it if something is wrong. Alternative suggestions are welcome.

Thank you,

--Jordan Atlas

Jordan,

No, this has been requested before but it is not an easy modification of mpl to make. The alternative suggestion is to always encapsulate the making of a figure in a script, and then save the script and the data. If you want to save the combination in a single file, then arrange for the script and data to reside in a single subdirectory, and zip or tar that subdirectory.

Eric

Can someone recomend a way to save the data in such a way that the columns (or rows) are labeled? In otherwords, it would be nice to be able to open the saved data and know what each row is without having to refer to the script that created it. (referring to the creating script feels error prone when you have many rows of data being saved). I'm currently using the 'pylab.save' function to save the data.

Thanks,

--Jordan

Eric Firing wrote:

···

Jordan Atlas wrote:

    Is it possible to 'save' a matplotlib figure object using something like the python pickle module? Basically, I'd like to save the 'figure' as a file so that I can open it and manipulate it if something is wrong. Alternative suggestions are welcome.

No, this has been requested before but it is not an easy modification of mpl to make. The alternative suggestion is to always encapsulate the making of a figure in a script, and then save the script and the data. If you want to save the combination in a single file, then arrange for the script and data to reside in a single subdirectory, and zip or tar that subdirectory.

I suggest using numpy record arrays for this -- the columns have names
and the data can be of different types. You can save and load them
using pickle (numpy.load and numpy.save will use pickle under the
hood).

If you want to stick with ASCII flat file representation (eg for use
with other programs), in matplotlib svn there are two nice functions
to help here: rec2csv and csv2rec. They support saving numpy record
arrays to CSV files with column names, and loading these back up later
doign type introspection to figure out the types (datetime, str,
float, int).

In [1]: import numpy as n

In [2]: import matplotlib.mlab as mlab

In [3]: x = n.random.rand(20,4)

In [4]: r = n.rec.fromrecords(x, names='age,weight,height,cash')

In [5]: r.dtype
Out[5]: dtype([('age', '<f8'), ('weight', '<f8'), ('height', '<f8'),
('cash', '<f8')])

In [7]: mlab.rec2csv(r, 'mydata.csv')

In [8]: !head mydata.csv
age,weight,height,cash
0.0449935,0.252057,0.316116,0.0635711
0.777189,0.155186,0.0537382,0.233598
0.731376,0.654577,0.977792,0.0171022
0.685975,0.373741,0.714592,0.620079
0.634548,0.956708,0.360962,0.885379
0.431011,0.359094,0.21484,0.961865
0.115155,0.78767,0.352753,0.769402
0.984747,0.720163,0.887608,0.316844
0.0478857,0.813668,0.882535,0.8837

In [9]: newr = mlab.csv2rec('mydata.csv')

In [10]: newr.dtype
Out[10]: dtype([('age', '<f8'), ('weight', '<f8'), ('height', '<f8'),
('cash', '<f8')])

In [11]: newr
Out[11]:
recarray([ (0.044993499999999999, 0.25205699999999998,
0.31611600000000001, 0.063571100000000005),
       (0.77718900000000002, 0.15518599999999999, 0.0537382, 0.233598),
       (0.73137600000000003, 0.65457699999999996, 0.97779199999999999,
0.017102200000000001),
       (0.685975, 0.37374099999999999, 0.714592, 0.62007900000000005),
       (0.634548, 0.956708, 0.36096200000000001, 0.88537900000000003),
       (0.43101099999999998, 0.35909400000000002, 0.21484, 0.96186499999999997),
       (0.11515499999999999, 0.78766999999999998, 0.35275299999999998,
0.76940200000000003),
       (0.98474700000000004, 0.720163, 0.88760799999999995,
0.31684400000000001),
       (0.047885700000000003, 0.81366799999999995,
0.88253499999999996, 0.88370000000000004),
       (0.044475599999999997, 0.89918900000000002,
0.076484499999999997, 0.114994),
       (0.75139299999999998, 0.70954300000000003, 0.458505,
0.33839900000000001),
       (0.14619299999999999, 0.907717, 0.24915200000000001,
0.67030400000000001),
       (0.89663199999999998, 0.61957300000000004,
0.0060039200000000003, 0.048883500000000003),
       (0.20794000000000001, 0.56046499999999999,
0.078303899999999996, 0.216032),
       (0.28726000000000002, 0.14282500000000001, 0.51740200000000003,
0.553037),
       (0.96326999999999996, 0.21327299999999999, 0.72040999999999999,
0.181446),
       (0.31984000000000001, 0.39338299999999998, 0.45787899999999998,
0.33919199999999999),
       (0.42086200000000001, 0.98801499999999998, 0.53429000000000004,
0.074105699999999997),
       (0.104211, 0.15845100000000001, 0.13339200000000001,
0.99228300000000003),
       (0.73563299999999998, 0.948407, 0.44708900000000001,
0.79521399999999998)],
      dtype=[('age', '<f8'), ('weight', '<f8'), ('height', '<f8'),
('cash', '<f8')])

You can also work with non floating point data

In [14]: url = ‘http://ichart.finance.yahoo.com/table.csv?s=GE&d=10&e=20&f=2007&g=d&a=0&b=2&c=1962&ignore=.csv

In [15]: import urllib

In [16]: urllib.urlretrieve(url, 'ge.csv')
Out[16]: ('ge.csv', <httplib.HTTPMessage instance at 0x8fa7b2c>)

In [17]: r = mlab.csv2rec('ge.csv')

In [18]: !head ge.csv
Date,Open,High,Low,Close,Volume,Adj Close
2007-11-19,38.48,38.51,38.00,38.16,35415000,38.16
2007-11-16,38.50,38.67,37.87,38.65,50181100,38.65
2007-11-15,38.93,38.93,38.13,38.31,41590000,38.31
2007-11-14,39.90,39.95,38.82,39.01,39650800,39.01
2007-11-13,38.50,39.25,38.25,39.21,42053400,39.21
2007-11-12,38.24,39.04,38.17,38.25,36968000,38.25
2007-11-09,38.52,38.75,38.11,38.38,42662200,38.38
2007-11-08,39.20,39.32,37.50,39.02,52970300,39.02
2007-11-07,39.90,39.93,38.99,39.08,46720100,39.08

In [20]: r[:10]
Out[20]:
recarray([ (datetime.datetime(2007, 11, 19, 0, 0), 38.479999999999997,
38.509999999999998, 38.0, 38.159999999999997, 35415000,
38.159999999999997),
       (datetime.datetime(2007, 11, 16, 0, 0), 38.5,
38.670000000000002, 37.869999999999997, 38.649999999999999, 50181100,
38.649999999999999),
       (datetime.datetime(2007, 11, 15, 0, 0), 38.93, 38.93,
38.130000000000003, 38.310000000000002, 41590000, 38.310000000000002),
       (datetime.datetime(2007, 11, 14, 0, 0), 39.899999999999999,
39.950000000000003, 38.82, 39.009999999999998, 39650800,
39.009999999999998),
       (datetime.datetime(2007, 11, 13, 0, 0), 38.5, 39.25, 38.25,
39.210000000000001, 42053400, 39.210000000000001),
       (datetime.datetime(2007, 11, 12, 0, 0), 38.240000000000002,
39.039999999999999, 38.170000000000002, 38.25, 36968000, 38.25),
       (datetime.datetime(2007, 11, 9, 0, 0), 38.520000000000003,
38.75, 38.109999999999999, 38.380000000000003, 42662200,
38.380000000000003),
       (datetime.datetime(2007, 11, 8, 0, 0), 39.200000000000003,
39.32, 37.5, 39.020000000000003, 52970300, 39.020000000000003),
       (datetime.datetime(2007, 11, 7, 0, 0), 39.899999999999999,
39.93, 38.990000000000002, 39.079999999999998, 46720100,
39.079999999999998),
       (datetime.datetime(2007, 11, 6, 0, 0), 40.200000000000003,
40.490000000000002, 39.969999999999999, 40.18, 42131000, 40.18)],
      dtype=[('date', '|O4'), ('open', '<f8'), ('high', '<f8'),
('low', '<f8'), ('close', '<f8'), ('volume', '<i4'), ('adj_close',
'<f8')])

···

On Nov 20, 2007 10:43 AM, Jordan Atlas <jca33@...163...> wrote:

Can someone recomend a way to save the data in such a way that the
columns (or rows) are labeled? In otherwords, it would be nice to be
able to open the saved data and know what each row is without having to
refer to the script that created it. (referring to the creating script
feels error prone when you have many rows of data being saved). I'm
currently using the 'pylab.save' function to save the data.