def make_ref(df, method='mean'):
''' Get a reference spectrum, requried for computing the dynamic spectrum (y-ref). Usually this is this is set
to the time-wise mean of the dataset, to 0, or to an external, pre-saved spectrum. This will generate mean or empy
reference spectra. External spectra are easy enough to generate.
Assumes spectral information is along the index of the dataframe! No attempt to acommadate other styles is made.
df: DataFrame with spectral data along index/row axis (=1) and temporal/physical variable along columns (axis=0)
Method: Style to generate reference spectrum from dataframe.
"mean" - Columnwise-mean of the dataframe
"empty" - Fills series with 0.0's to length of spectral index
returns: series of length of df.index'''
method=method.lower()
if method=='mean':
refspec=df.mean(axis=1)
elif method=='empty':
refspec=Series( [0.0 for i in range(len(df.index))], index=df.index) #builtin way to do this?
else:
raise badvalue_error(method, 'mean, empty')
refspec.name='refspec' #Not sure if this will be useful
return refspec
def divby(df, divisor=0, axis=0, sameshape=True):
''' Small wrapper around pandas.divide() that takes in either column integers, names,
or entire arrays as the divisor. Note, pandas.divide by default is rowwise, but here
it defaults to column-wise.
Parameters:
-----------
df: Dataframe object.
divisor: (See notes below)
axis: Division axis. 0=column-wise division. 1=row-wise
sameshape: If true, function will check that the divisor as the same shape as
the data along that axis. For example, is an array of 2000 units,
and axis=0, then the row dimension must be 2000 for the dataframe.
AFAIK df.div does not enforce this stringency, which sometimes can lead
to hard-to-find bugs.
Notes:
------
This program first checks if divisor is an integer. If not, it attempts
to lookup by rowname or columnname. Finally, it tries to divide by the divisor itself,
if the divisor is an array/series. This seemed like a flexible way to do this
without too much type checking.'''
### Column-wise division
if axis==0:
if isinstance(divisor, int):
divisor=df.columns[divisor]
else:
try:
divisor=df[divisor]
except Exception:
pass #divisor=divisor
divlen=df.shape[0] #Store length of a single column
### Rowwise division
elif axis==1:
if isinstance(divisor, int):
divisor=df.ix[divisor]
else:
try:
divisor=df.xs(divisor)
except KeyError:
pass
divlen=df.shape[1]
else:
raise badvalue_error('axis', '0,1')
### Enforce strict shapetype
if sameshape:
if len(divisor) != divlen:
raise TypeError('Divisor dimensions %s do not match dividend'
' dimension %s along axis = %s'%(divlen, len(divisor), axis))
return df.divide(divisor, axis=axis)
def divby(df, divisor=0, axis=0, sameshape=True):
''' Small wrapper around pandas.divide() that takes in either column integers, names,
or entire arrays as the divisor. Note, pandas.divide by default is rowwise, but here
it defaults to column-wise.
Parameters:
-----------
df: Dataframe object.
divisor: (See notes below)
axis: Division axis. 0=column-wise division. 1=row-wise
sameshape: If true, function will check that the divisor as the same shape as
the data along that axis. For example, is an array of 2000 units,
and axis=0, then the row dimension must be 2000 for the dataframe.
AFAIK df.div does not enforce this stringency, which sometimes can lead
to hard-to-find bugs.
Notes:
------
This program first checks if divisor is an integer. If not, it attempts
to lookup by rowname or columnname. Finally, it tries to divide by the divisor itself,
if the divisor is an array/series. This seemed like a flexible way to do this
without too much type checking.'''
### Column-wise division
if axis == 0:
if isinstance(divisor, int):
divisor = df.columns[divisor]
else:
try:
divisor = df[divisor]
except Exception:
pass #divisor=divisor
divlen = df.shape[0] #Store length of a single column
### Rowwise division
elif axis == 1:
if isinstance(divisor, int):
divisor = df.ix[divisor]
else:
try:
divisor = df.xs(divisor)
except KeyError:
pass
divlen = df.shape[1]
else:
raise badvalue_error('axis', '0,1')
### Enforce strict shapetype
if sameshape:
if len(divisor) != divlen:
raise TypeError('Divisor dimensions %s do not match dividend'
' dimension %s along axis = %s' %
(divlen, len(divisor), axis))
return df.divide(divisor, axis=axis)
def cmget(color):
''' Return colormap from string (eg 'jet')'''
try:
cmap=getattr(cm, color)
except AttributeError:
raise badvalue_error(color, 'a custom LInearSegmentedColormap or a string matching \
the default colormaps available to matplotlib (http://dept.astro.lsa.umich.edu/~msshin/science/code/matplotlib_cm/)')
return cmap
def _df_colormapper(df, cmap, axis=0, colorbymax=False, vmin=None, vmax=None):
''' Maps matplotlibcolors to a dataframe based on the mean value of each curve along that
axis.
Parameters
----------
colorbymax : bool (False)
If true, curves are colored based on their maxima value (ie largest
curve is max color regardless of where it appears in time). Otherwise,
curves are colored chornologically.
Notes
-----
Useful for df.plot() which doesn't take a normalized colormap natively.
cmap can be an instance of an RGB color map, or a string which such that
cm.string will produce one.
'''
if isinstance(cmap, basestring):
cmap = cmget(cmap)
if axis != 0 and axis != 1:
raise badvalue_error(axis, 'integers 0 or 1')
# Min and max values of color map must be min and max of dataframe
if not vmin:
try:
vmin = min(df.min(axis=axis))
except TypeError: #If df only is one column...
vmin = df.min(axis=axis)
if not vmax:
try:
vmax = max(df.max(axis=axis))
except TypeError:
vmax = df.max(axis=axis)
cNorm = mplcolors.Normalize(vmin=vmin, vmax=vmax)
scalarmap = cm.ScalarMappable(norm=cNorm, cmap=cmap)
if axis == 0:
if colorbymax:
colors = [scalarmap.to_rgba(df[x].max()) for x in df.columns]
else:
colors = [
scalarmap.to_rgba(x)
for x in np.linspace(vmin, vmax, len(df.columns))
]
elif axis == 1:
if colorbymax:
colors = [scalarmap.to_rgba(df.ix[x].max()) for x in df.index]
else:
colors = [
scalarmap.to_rgba(x)
for x in np.linspace(vmin, vmax, len(df.index))
]
return colors
def cmget(color):
''' Return colormap from string (eg 'jet')'''
try:
cmap = getattr(cm, color)
except AttributeError:
raise badvalue_error(
color, 'a custom LInearSegmentedColormap or a string matching \
the default colormaps available to matplotlib (http://dept.astro.lsa.umich.edu/~msshin/science/code/matplotlib_cm/)'
)
return cmap
def _df_colormapper(df, cmap, axis=0, colorbymax=False, vmin=None, vmax=None):
''' Maps matplotlibcolors to a dataframe based on the mean value of each curve along that
axis.
Parameters
----------
colorbymax : bool (False)
If true, curves are colored based on their maxima value (ie largest
curve is max color regardless of where it appears in time). Otherwise,
curves are colored chornologically.
Notes
-----
Useful for df.plot() which doesn't take a normalized colormap natively.
cmap can be an instance of an RGB color map, or a string which such that
cm.string will produce one.
'''
if isinstance(cmap, basestring):
cmap=cmget(cmap)
if axis != 0 and axis != 1:
raise badvalue_error(axis, 'integers 0 or 1')
# Min and max values of color map must be min and max of dataframe
if not vmin:
try:
vmin = min(df.min(axis=axis))
except TypeError: #If df only is one column...
vmin = df.min(axis=axis)
if not vmax:
try:
vmax = max(df.max(axis=axis))
except TypeError:
vmax = df.max(axis=axis)
cNorm = mplcolors.Normalize(vmin=vmin, vmax=vmax)
scalarmap = cm.ScalarMappable(norm=cNorm, cmap=cmap)
if axis == 0:
if colorbymax:
colors=[scalarmap.to_rgba(df[x].max()) for x in df.columns]
else:
colors = [scalarmap.to_rgba(x) for x in
np.linspace(vmin, vmax, len(df.columns))]
elif axis == 1:
if colorbymax:
colors=[scalarmap.to_rgba(df.ix[x].max()) for x in df.index]
else:
colors = [scalarmap.to_rgba(x) for x in
np.linspace(vmin, vmax, len(df.index))]
return colors
def _annotate_mappable(df, cmap, axis=0, vmin=None, vmax=None):
if isinstance(cmap, basestring):
cmap = cmget(cmap)
if axis != 0 and axis != 1:
raise badvalue_error(axis, 'integers 0 or 1')
# Min and max values of color map must be min and max of dataframe
if not vmin:
vmin = min(df.min(axis=axis))
if not vmax:
vmax = max(df.max(axis=axis))
cNorm = mplcolors.Normalize(vmin=vmin, vmax=vmax)
scalarmap = cm.ScalarMappable(norm=cNorm, cmap=cmap)
# http://stackoverflow.com/questions/6600579/colorbar-for-matplotlib-plot-surface-command
scalarmap.set_array(np.arange(0, 1)) #Thsi can be anything, arbitrary
return scalarmap, vmin, vmax
def _annotate_mappable(df, cmap, axis=0, vmin=None, vmax=None):
if isinstance(cmap, basestring):
cmap=cmget(cmap)
if axis != 0 and axis != 1:
raise badvalue_error(axis, 'integers 0 or 1')
# Min and max values of color map must be min and max of dataframe
if not vmin:
vmin=min(df.min(axis=axis))
if not vmax:
vmax=max(df.max(axis=axis))
cNorm = mplcolors.Normalize(vmin=vmin, vmax=vmax)
scalarmap = cm.ScalarMappable(norm=cNorm, cmap=cmap)
# http://stackoverflow.com/questions/6600579/colorbar-for-matplotlib-plot-surface-command
scalarmap.set_array(np.arange(0,1)) #Thsi can be anything, arbitrary
return scalarmap, vmin, vmax
def easy_legend(ax, fancy=True, position='top', **legendkwds):
''' Wrapper around ax.legend to make it easy to move a legend around the edges of the plot. Made for sensible
numbers of lines (1-15) and tries to choose smart placement to avoid conflict with lines, labels etc...
BROKEN!!!
This is a bust, since plotting legend is easy enough. See guide, especially with the 'loc'
http://matplotlib.org/users/legend_guide.html
If coming back to this, here are the issues to resolve:
legend['loc'] must be enabled for the bounding box to work correctly/consistently.
bbox coordinates are (left edge, bottom edge) of legend.
-controlling bottom edge is really dumb because want the top edge to clear the xlabel for example, and bottom
edge depends on thickness of plot. Putting the legend on top actually works nicely for this.
If plotting outside left/right of plot, need to squeeze the plot in afterall, it will not accommadate my legend.
Had this code, but deleted it. Basically, need to squeeze left/right width of the axes.bounding width by 20%
per column of legend.
(see second reply here http://stackoverflow.com/questions/4700614/how-to-put-the-legend-out-of-the-plot)
'''
ir = lambda x: int(round(x))
position = position.lower()
legendkwds[
'loc'] = 3 #MUST BE SET TO SOMETHING, 2 or 3 doesn't matter, or lower left is no longer start point
if position not in ['top', 'bottom', 'left', 'right']:
raise badvalue_error(position, 'top, bottom, left, or right')
################################################################
### Choose skinny legend for top bottom, long for left/right ###
################################################################
if 'ncol' not in legendkwds:
if position == 'top' or position == 'bottom':
if len(ax.lines) < 4:
ncol = len(ax.lines)
else:
ncol = 4
else:
ncol = ir(
len(ax.lines) / 20.0
) #For left right, nice if columns have large vertical extent (10 lines is good)
if ncol == 0:
ncol = 1
legendkwds['ncol'] = ncol
###################################################
### Choose legend position around plot elements ###
###################################################
box = ax.get_position(
) #Can do box.x0 and stuff when resetting ax.set_position()
if 'bbox_to_anchor' not in legendkwds: #Don't use has_key(), gone in 3.x
if position == 'top':
''' Legend slightly inside the upper bound of plot'''
if ax.get_title() == '':
bbox_to_anchor = (
0.2, 1.025
) #0.25 will only center plot if plot is 0.5 units long, then
else: #other end at 0.75. 0.2 looks nice for 8 column uv-vis plot.
bbox_to_anchor = (
0.2, 1.05
) #Also depends on size of monitor!! so either way, screwed
elif position == 'bottom':
''' Centered legend under the label'''
if ax.get_xlabel() == '':
bbox_to_anchor = (0.25, -0.025)
else:
bbox_to_anchor = (0.25, -0.05)
elif position == 'right':
''' Squeeze 20% width inward per column in legend to make room'''
bbox_to_anchor = (1.05, 0.0) #WHY DOESNT 1,1 work
elif position == 'left':
''' Squeeze 20% width inward per column in legend to make room'''
if ax.get_ylabel() == '':
bbox_to_anchor = (-0.07, 1.0)
else:
bbox_to_anchor = (-0.12, 1.0)
legendkwds['bbox_to_anchor'] = bbox_to_anchor
if fancy and 'fancybox' not in legendkwds and 'shadow' not in legendkwds:
legendkwds['fancybox'] = True
legendkwds['shadow'] = True
if 'borderaxespad' not in legendkwds:
legendkwds['borderaxespad'] = 0.0 #Havne't played with this
### WHY IS MODE BROKEN (comes up garbled on plot)
# if 'mode' not in legendkwds:
# legendkwds['mode']='expand'
ax.legend(**legendkwds)
return ax
def _gen2d(xx, yy, zz, contours=6, label=None, colorbar=None,
background=None, **pltkwargs):
""" Abstract layout for 2d plot.
For convienence, a few special labels, colorbar and background keywords
have been implemented. If these are not adequate, it one can add
custom colorbars, linelabels background images etc... easily just by
using respecitve calls to plot (plt.colorbar(), plt.imshow(),
plt.clabel() ); my implementations are only for convienences and
some predefined, cool styles.
countours: Number of desired contours from output.
label: Predefined label types. For now, only integer values 1,2. Use plt.clabel to add a custom label.
background: Integers 1,2 will add gray or autumn colormap under contour plot. Use plt.imgshow() to generate
custom background, or pass a PIL-opened image (note, proper image scaling not yet implemented).
fill: bool (False)
Fill between contour lines.
**pltkwargs: Will be passed directly to plt.contour().
"""
# Boilerplate from basic_plots._genplot(); could refactor as decorator
xlabel = pltkwargs.pop('xlabel', '')
ylabel = pltkwargs.pop('ylabel', '')
title = pltkwargs.pop('title', '')
grid = pltkwargs.pop('grid', '')
fill = pltkwargs.pop('fill', True)
# pltkwargs.setdefault('legend', False) #(any purpose in 2d?)
pltkwargs.setdefault('linewidth', 1)
cbar = pltkwargs.pop('cbar', False)
fig = pltkwargs.pop('fig', plt.gcf())
ax = pltkwargs.pop('ax', None)
# Overwrites fig (desireable?)
if not ax:
fig, ax = plt.subplots(1)
labelsize = pltkwargs.pop('labelsize', 'medium') #Can also be ints
titlesize = pltkwargs.pop('titlesize', 'large')
ticksize = pltkwargs.pop('ticksize', '') #Put in default and remove bool gate
### BUG, PLT.CONTOUR() doesn't take 'color'; rather, takes 'colors' for now
### This is only for solid colors. contour takes cmap argument by default.
if 'color' in pltkwargs:
pltkwargs['colors']=pltkwargs.pop('color')
### Convienence method to pass in string colors
if 'cmap' in pltkwargs and isinstance(pltkwargs['cmap'], basestring):
pltkwargs['cmap']=pu.cmget(pltkwargs['cmap'])
### I CAN MAKE THIS GENERAL TO ANY PLOT, ADDITION OF IMAGES TO A BACKGROUND MAKE IT A GENERAL ROUTINE TO ALL MY PLOTS ###
### More here http://matplotlib.org/examples/pylab_examples/image_demo3.html ###
# Refactored with xx, yy instead of df.columns/index UNTESTED
if background:
xmin, xmax, ymin, ymax = xx.min(), xx.max(), yy.min(), yy.max()
if background==1:
im = plt.imshow(zz, interpolation='bilinear', origin='lower',
cmap=cm.gray, extent=(xmin, xmax, ymin, ymax))
elif background==2:
im = plt.imshow(zz, interpolation='bilinear', origin='lower',
cmap=cm.autumn, extent=(xmin, xmax, ymin, ymax))
### This will take a custom image opened in PIL or it will take plt.imshow() returned from somewhere else
else:
try:
im = plt.imshow(background)
### Perhaps image was not correctly opened
except Exception:
raise badvalue_error(background, 'integer 1,2 or a PIL-opened image')
else:
im=None
if fill:
contours = ax.contourf(xx, yy, zz, contours, **pltkwargs) #linewidths is a pltkwargs arg
else:
contours = ax.contour(xx, yy, zz, contours, **pltkwargs)
if cbar:
fig.colorbar(contours)
### Pick a few label styles to choose from.
if label:
if label==1:
ax.clabel(inline=1, fontsize=10)
elif label==2:
ax.clabel(levels[1::2], inline=1, fontsize=10) #label every second line
else:
raise PlotError(label, 'integer of value 1 or 2')
ax.set_xlabel(xlabel, fontsize=labelsize)
ax.set_ylabel(ylabel, fontsize=labelsize)
ax.set_title(title, fontsize=titlesize)
if grid:
ax.grid()
return (ax, contours)
def plot3d(df, kind='contour', elev=0, azim=0, proj_xy=True, proj_zy=True, proj_xz=True,
contour_color=None, contour_cmap=None, c_iso=10, r_iso=10,*args, **pltkwargs):
""" Matplotlib Axes3d wrapper for dataframe. Made to handle surface plots, so pure contour plots,
polygon plots etc... these need their own classes.
parameters:
kind
contour: Based on Axes3d.contour(x,y,z,*args,**kwargs), returns a plot_surface() with
relevant spectral projections.
contourf: Filled contour plot based on Axes3D.conourf(x,y,z,*args,**kwargs). Same a contour
except contours are filled.
pltwargs can be any Axes3d keywords, with the following extra keywords being germane to this class.
xlabel, ylabel, zlabel: Axis labels. Defaults are provided here. If none are provided, the
attributes df.index.name and df.columns.name will be examined, aftwards,
defaults are used (see below).
xlim, ylim, zlim: These can be passed, but if not, the are inferred from the data. It's smarter
to manipulate the df input than to mess with these. For now, I'm keeping them
because they are relevant to proj_padding, but otherwise, probably not too essential.
proj_padding: Padding percentage if one does not want contour projections right on the wall of the grid.
This actually needs fixed. It is intended to work as in the user says... 1.1 and this means
that the projections should be 10% outside of the axes grid. 0.9 would be 10% in.
The operation to computes the new limit is actually not correct, and if a minimum axis
value is 0, it won't move it beyond.
If this ever becomes useful again, I think it would be wise to first set the x,y,z lims
on the plot, then after that, compute the paddings and set the offset to these values.
Something like
ax.set(xlim)
newxlim=xlim*padding correction
contour(offset=newxlim)
A note on strides: Strides are step sizes. Say you have 100 curves, then a stride of 5 would give you 20
isolines over your surface. As such, a smart keyword (c_iso and r_iso) will make sure
that a fixed number of contours are drawn when the user passes in a dataframe.
c_iso, r_iso: See above. These control how man column and row iso lines will be projected onto the 3d plot.
For example, if c_iso=10, then 10 isolines will be plotted as columns, despite the actual length of the
columns. Alternatively, one can pass in c_stride directly, which is a column step size rather than
an absolute number, and c_iso will be disregarded.
contour_color/contour_cmap:
Either a solid color, or colormap, passed to the contour projections (proj_xy etc...).
For now, will be passed to all contour projections (xy, zy, xz).
"""
### Error raised by cfunc() is not very clear so here's an easier one
if contour_cmap and contour_color:
raise AttributeError('plot3d can have non-null values for attributes contour_cmap and contour_color, but not both.')
zlabel_def=''
zlabel=pltkwargs.pop('zlabel', zlabel_def)
xlabel = pltkwargs.pop('xlabel', '')
ylabel = pltkwargs.pop('ylabel', '')
title = pltkwargs.pop('title', '')
### If plane (xy) is input backwards (yx), still works
proj_xy=pltkwargs.pop('proj_yx', proj_xy)
proj_zy=pltkwargs.pop('proj_yz', proj_zy)
proj_xz=pltkwargs.pop('proj_zx', proj_xz)
### Make mesh grid based on dataframe indicies ###
fig = plt.figure()
ax = fig.gca(projection='3d')
xx,yy=np.meshgrid(df.columns, df.index)
### Set axis limits explicitly from padding and data.
### PLOT PADDING NEEDS FIXED AS OF NOW IT IS NOT CORRECT. FOR EXAMPLE, IF MIN IS 0, NEED AN OPERATION
### THAT WILL STILL REDUCE IT TO LESS THAN ZERO. ONLY WORTH FIGURING OUT IF NEED PROJECTIONS OUTSIDE OF THE GRID.
### ALSO BUGS IF THE AXIS VALUES AREN'T NUMBERS. FOR EXAMPLE, IF THEY ARE NAMED COLUMNS. FIX WITH TRY EXCEPT
proj_padding=pltkwargs.pop('projpad', 0.0)
xlim=pltkwargs.pop('xlim', (min(df.columns)*(1.0-proj_padding), max(df.columns)*(1.0+proj_padding)) )
ylim=pltkwargs.pop('ylim', (min(df.index)*(1.0-proj_padding), max(df.index)*(1.0+proj_padding)) )
### Min/max of total data (only works if the data is 2d)
zlim=pltkwargs.pop('zlim', ((1.0-proj_padding)*(df.min()).min(), (1.0+proj_padding)*(df.max().max())) )
cstride= _ir( len(df.columns)/float(c_iso) )
rstride= _ir( len(df.index)/float(r_iso) )
### This occurs if the
if cstride==0:
print "Warning, dataset is too small to accomodate c_iso of %i, setting to 1."%c_iso
cstride=1
if rstride==0:
print "Warning, dataset is too small to accomodate r_iso of %i, setting to 1."%r_iso
rstride=1 #to prevent errors
### If these values are already passed in, they won't be overwritten. Cmap here is for surface, not countour
_surface_defaults={'alpha':0.2, 'rstride':rstride, 'cstride':cstride, 'cmap':pu.cmget('autumn')} #Don't have any special
_surface_defaults.update(pltkwargs)
pltkwargs=_surface_defaults #Need to do this in two steps or it errors
ax.plot_surface(xx,yy,df, *args, **pltkwargs)
if kind=='contourf' or kind=='contour':
if kind=='contourf':
cfunc=ax.contourf
else:
cfunc=ax.contour
if proj_xy:
cset = cfunc(xx, yy, df, zdir='z', colors=contour_color, cmap=contour_cmap, offset=zlim[0]) #project z onto xy (zmid to bottom)
if proj_zy:
cset = cfunc(xx, yy, df, zdir='x',colors=contour_color, cmap=contour_cmap, offset=xlim[0]) #project x onto zy (timestart) (negative half time interval)
if proj_xz:
cset = cfunc(xx, yy, df, zdir='y',colors=contour_color, cmap=contour_cmap, offset=ylim[1]) #project y onto xz (ymid to 0) (negative mid wavelength)
else:
raise badvalue_error(kind, 'contour, contourf')
ax.set_xlabel(xlabel) #x
ax.set_ylabel(ylabel) #Y
ax.set_zlabel(zlabel) #data
ax.set_title(title)
ax.view_init(elev, azim)
return ax
def plot2d(df, contours=6, label=None, colorbar=None, background=None, **pltkwds):
''' Wrapper for plt.contour that uses np.meshgrid to generate curves. For convienence, a few special labels, colorbar and background
keywords have been implemented. If these are not adequate, it one can add custom colorbars, linelabels background images etc... easily
just by using respecitve calls to plot (plt.colorbar(), plt.imshow(), plt.clabel() ); my implementations are only for convienences and
some predefined, cool styles.
df: Dataframe.
countours: Number of desired contours from output.
label: Predefined label types. For now, only integer values 1,2. Use plt.clabel to add a custom label.
background: Integers 1,2 will add gray or autumn colormap under contour plot. Use plt.imgshow() to generate
custom background, or pass a PIL-opened image (note, proper image scaling not yet implemented).
**pltkwds: Will be passed directly to plt.contour().'''
xlabel, ylabel, title, pltkwds=pu.smart_label(df, pltkwds)
### BUG, PLT.CONTOUR() doesn't take 'color'; rather, takes 'colors' for now
### This is only for solid colors. contour takes cmap argument by default.
if 'color' in pltkwds:
pltkwds['colors']=pltkwds.pop('color')
### Convienence method to pass in string colors
if 'cmap' in pltkwds and isinstance(pltkwds['cmap'], basestring):
pltkwds['cmap']=pu.cmget(pltkwds['cmap'])
### I CAN MAKE THIS GENERAL TO ANY PLOT, ADDITION OF IMAGES TO A BACKGROUND MAKE IT A GENERAL ROUTINE TO ALL MY PLOTS ###
### More here http://matplotlib.org/examples/pylab_examples/image_demo3.html ###
if background:
xmin, xmax, ymin, ymax=min(df.columns), max(df.columns), min(df.index), max(df.index)
if background==1:
im = plt.imshow(df, interpolation='bilinear', origin='lower',
cmap=cm.gray, extent=(xmin, xmax, ymin, ymax))
elif background==2:
im = plt.imshow(df, interpolation='bilinear', origin='lower',
cmap=cm.autumn, extent=(xmin, xmax, ymin, ymax))
### This will take a custom image opened in PIL or it will take plt.imshow() returned from somewhere else
else:
try:
im = plt.imshow(background)
### Perhaps image was not correctly opened
except Exception:
raise badvalue_error(background, 'integer 1,2 or a PIL-opened image')
else:
im=None
xx,yy = np.meshgrid(df.columns, df.index)
ax = plt.contour(xx, yy, df, contours, **pltkwds) #linewidths is a pltkwds arg
### Pick a few label styles to choose from.
if label:
if label==1:
ax.clabel(inline=1, fontsize=10)
elif label==2:
ax.clabel(levels[1::2], inline=1, fontsize=10) #label every second line
else:
raise badvalue_error(label, 'integer of value 1 or 2')
### Add colorbar, for some reason need to do plt.colorbar() even though can do ax.clabel...
if colorbar:
if colorbar==1: ### WHAT OTHER COLORBAR OPTIONS ARETHERE
CB=plt.colorbar(ax, shrink=0.8) #shringk is heigh of colorbar relative ot height of plot
if im:
IMCB=plt.colorbar(im, orientation='horizontal', shrink=0.8)
### Move original colorbar to more natural to make room for image colorbar
l,b,w,h = plt.gca().get_position().bounds
ll,bb,ww,hh = CB.ax.get_position().bounds
CB.ax.set_position([ll, b+0.1*h, ww, h*0.8])
else:
raise badvalue_error(colorbar, 'integer of value 1 is only supported for now')
plt.xlabel(xlabel)
plt.ylabel(ylabel)
plt.title(title)
return ax
def easy_legend(ax, fancy=True, position='top', **legendkwds):
''' Wrapper around ax.legend to make it easy to move a legend around the edges of the plot. Made for sensible
numbers of lines (1-15) and tries to choose smart placement to avoid conflict with lines, labels etc...
BROKEN!!!
This is a bust, since plotting legend is easy enough. See guide, especially with the 'loc'
http://matplotlib.org/users/legend_guide.html
If coming back to this, here are the issues to resolve:
legend['loc'] must be enabled for the bounding box to work correctly/consistently.
bbox coordinates are (left edge, bottom edge) of legend.
-controlling bottom edge is really dumb because want the top edge to clear the xlabel for example, and bottom
edge depends on thickness of plot. Putting the legend on top actually works nicely for this.
If plotting outside left/right of plot, need to squeeze the plot in afterall, it will not accommadate my legend.
Had this code, but deleted it. Basically, need to squeeze left/right width of the axes.bounding width by 20%
per column of legend.
(see second reply here http://stackoverflow.com/questions/4700614/how-to-put-the-legend-out-of-the-plot)
'''
ir=lambda x: int(round(x))
position=position.lower()
legendkwds['loc']=3 #MUST BE SET TO SOMETHING, 2 or 3 doesn't matter, or lower left is no longer start point
if position not in ['top', 'bottom', 'left', 'right']:
raise badvalue_error(position, 'top, bottom, left, or right')
################################################################
### Choose skinny legend for top bottom, long for left/right ###
################################################################
if 'ncol' not in legendkwds:
if position=='top' or position=='bottom':
if len(ax.lines) < 4:
ncol=len(ax.lines)
else:
ncol=4
else:
ncol=ir( len(ax.lines)/20.0) #For left right, nice if columns have large vertical extent (10 lines is good)
if ncol==0:
ncol=1
legendkwds['ncol']=ncol
###################################################
### Choose legend position around plot elements ###
###################################################
box=ax.get_position() #Can do box.x0 and stuff when resetting ax.set_position()
if 'bbox_to_anchor' not in legendkwds: #Don't use has_key(), gone in 3.x
if position=='top':
''' Legend slightly inside the upper bound of plot'''
if ax.get_title() == '':
bbox_to_anchor=(0.2, 1.025) #0.25 will only center plot if plot is 0.5 units long, then
else: #other end at 0.75. 0.2 looks nice for 8 column uv-vis plot.
bbox_to_anchor=(0.2, 1.05) #Also depends on size of monitor!! so either way, screwed
elif position=='bottom':
''' Centered legend under the label'''
if ax.get_xlabel()=='':
bbox_to_anchor=(0.25, -0.025)
else:
bbox_to_anchor=(0.25, -0.05)
elif position=='right':
''' Squeeze 20% width inward per column in legend to make room'''
bbox_to_anchor=(1.05, 0.0) #WHY DOESNT 1,1 work
elif position=='left':
''' Squeeze 20% width inward per column in legend to make room'''
if ax.get_ylabel()=='':
bbox_to_anchor=(-0.07, 1.0)
else:
bbox_to_anchor=(-0.12, 1.0)
legendkwds['bbox_to_anchor']=bbox_to_anchor
if fancy and 'fancybox' not in legendkwds and 'shadow' not in legendkwds:
legendkwds['fancybox']=True
legendkwds['shadow']=True
if 'borderaxespad' not in legendkwds:
legendkwds['borderaxespad']=0.0 #Havne't played with this
### WHY IS MODE BROKEN (comes up garbled on plot)
# if 'mode' not in legendkwds:
# legendkwds['mode']='expand'
ax.legend(**legendkwds)
return ax