import datetime

import warnings

import numpy as np

import matplotlib

from hapiplot.plot.datetick import datetick

# https://github.com/pandas-dev/pandas/issues/18301

# Suppresses depreciation warning.

# TODO: determine what version of pandas this is needed for.

# Observed in Matplotlib 3.0, pandas 0.25.3, Python 3.5.

from pandas.plotting import register_matplotlib_converters

register_matplotlib_converters()

def timeseries(t, y, **kwargs):

"""Plot a time series"""

opts = {

'logging': False,

'title': '',

'xlabel': '',

'ylabel': '',

'logx': '',

'logy': '',

'nodata': False,

'backend': 'default',

'returnimage': False,

'transparent': False,

'legendlabels': []

}

for key, value in kwargs.items():

if key in opts:

opts[key] = value

else:

warnings.warn('Warning: Ignoring invalid keyword option "%s".' % key, SyntaxWarning)

if opts['returnimage']:

# When returnimage=True, the Matplotlib OO API is used b/c it is thread safe.

# Otherwise, the pyplot API is used. Ideally would always use the OO API,

# but this has problems with notebooks and showing figure when executing

# a script from the command line.

# TODO: Document issues.

#

# API differences description:

# https://www.saltycrane.com/blog/2007/01/how-to-use-pylab-api-vs-matplotlib-api_3134/

from matplotlib.figure import Figure

from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas

if opts['logging']:

from matplotlib import __version__ as mpl_version

print("timeseries(): Using Matplotlib %s with FigCanvasAgg back-end" % mpl_version)

else:

from matplotlib import pyplot as plt

if opts['backend'] != 'default':

try:

matplotlib.use(opts['backend'], force=True)

except:

matplotlib.use(matplotlib.get_backend(), force=True)

warnings.warn('Warning: matplotlib(' + opts['backend'] + \

') call failed. Using default backend of ' +

matplotlib.get_backend(), SyntaxWarning)

if opts['logging']:

from matplotlib import __version__ as mpl_version

print("timeseries(): Using Matplotlib %s with %s back-end" % (mpl_version, matplotlib.get_backend()))

if not isinstance(y, np.ndarray) and len(y) > 1 and len(y[0] > 1):

y = np.array(y).T

else:

y = np.array(y)

t = np.array(t)

if y.shape[0] != t.shape[0]:

if len(y.shape) > 1:

if y.shape[1] == t.shape[0]:

y = y.T

if len(t.shape) > 1 and y.shape[0] == t.shape[1]:

y = y.T

if y.shape[0] < 11:

props = {'linestyle': 'none', 'marker': '.', 'markersize': 16}

elif y.shape[0] < 101:

props = {'linestyle': '-', 'linewidth': 2, 'marker': '.', 'markersize': 8}

else:

props = {}

ylabels = []

if issubclass(y.dtype.type, np.flexible):

# See https://docs.scipy.org/doc/numpy-1.13.0/reference/arrays.scalars.html

# for diagram of subclasses.

# Find unique strings and give each an integer value.

# Modify tick labels to correspond to unique strings

yu = np.unique(y)

if len(yu) > 20:

# Too many labels in this case. One option is to find

# number of unique first characters and change labels to

# "first character" and then warn. If number of unique first

# characters < 21, try number of unique second characters, etc.

raise ValueError('Can only plot strings if number of unique strings < 21')

yi = np.zeros((y.shape))

for i in range(0, len(yu)):

yi[y == yu[i]] = i

ylabels = yu

y = yi

# Can't use matplotlib.style.use(style) because not thread safe.

# Set context using 'with'.

# Setting stylesheet method: https://stackoverflow.com/a/22794651/1491619

if opts['returnimage']:

# See note above about OO API for explanation for why this is

# done differently if returnimage=True

fig = Figure()

# Attach canvas to fig, which is needed by datetick and hapiplot.

FigureCanvas(fig)

ax = fig.add_subplot(111)

else:

fig, ax = plt.subplots()

if len(y.shape) > 1:

all_nan = np.full((y.shape[1]), False)

for i in range(0, y.shape[1]):

try:

all_nan[i] = np.all(np.isnan(y[:,i]))

except:

all_nan[i] = False

else:

all_nan = np.array([False])

try:

all_nan[0] = np.all(np.isnan(y))

except:

all_nan[0] = False

legendlabels = opts['legendlabels'].copy()

if legendlabels == []:

if len(y.shape) > 1:

for i in range(0, y.shape[1]):

legendlabels.append('col #{}'.format(i))

if np.any(all_nan):

if legendlabels != []:

if len(y.shape) > 1:

for i in range(0, y.shape[1]):

if all_nan[i] == True and opts['nodata'] == False:

legendlabels[i] = '{0:s}: All {1:d} values are NaN'.format(legendlabels[i], y.shape[0])

else:

legendlabels[i] = '{0:s}: No data in interval'.format(legendlabels[i])

else:

if opts['nodata'] == True:

legendlabels[0] = '{0:s}: No data in interval'.format(legendlabels[0])

else:

legendlabels[0] = '{0:s}: All {1:d} values are NaN'.format(legendlabels[0], y.shape[0])

else:

if len(y.shape) > 1:

for i in range(0, y.shape[1]):

if all_nan[i] == True and opts['nodata'] == False:

legendlabels[i] = 'All {0:d} values are NaN'.format(y.shape[0])

else:

legendlabels[i] = 'No data in interval'

else:

if opts['nodata'] == True:

legendlabels = ['No data in interval']

else:

legendlabels = ['All {0:d} values are NaN'.format(len(y))]

if np.all(all_nan):

ax.set_yticklabels([])

ax.set_yticks([])

if len(y.shape) == 1 and y.size == 1:

# Single time value. Set one tick having that value.

ax.set_yticks(y)

if np.any(all_nan):

if len(y.shape) > 1:

for i in range(0, y.shape[1]):

if all_nan[i]:

ax.plot([t[0],t[-1]],[0,0], alpha=0)

else:

ax.plot(t, y[:,i])

else:

ax.plot([t[0],t[-1]],[0,0], linestyle=None, alpha=0)

else:

ax.plot(t, y, **props)

ax.set(ylabel=opts['ylabel'], xlabel=opts['xlabel'], title=opts['title'])

try:

ax.ticklabel_format(axis='y', style='sci', scilimits=(-3,3), useMathText=True)

except:

pass

if legendlabels != []:

leg = fig.legend(legendlabels)

if np.any(all_nan):

for i in range(0, len(all_nan)):

leg.get_lines()[i].set_alpha(1)

ax.set_position([0.12,0.125,0.850,0.75])

if np.all(all_nan):

ax.grid(which='major', axis='x')

else:

ax.grid()

if not np.all(all_nan) and len(ylabels) > 0:

ax.set_yticks(np.unique(y))

ax.set_yticklabels(ylabels)

if isinstance(t[0], datetime.datetime):

datetick('x', axes=ax)

if isinstance(y[0], datetime.datetime):

datetick('y', axes=ax)

# savefig.transparent=True requires the following for the saved image

# to have a transparent background. Seems as though figure.facealpha

# and axes.facealpha should be rc parameters, but they are not. So

# savefig.transparent controls both transparency in saved image and

# in GUI image.

if opts['transparent']:

fig.patch.set_alpha(0)

ax.patch.set_alpha(0)

if not opts['returnimage']:

plt.show()

return fig

def adjust_labels(ax):

# Not used. See

# https://stackoverflow.com/questions/24581194/matplotlib-text-bounding-box-dimensions

# for determining text bounding box in figure coordinates

for item in ax.get_yticklabels():

ml = 0 # max length

for t in item.get_text().split('\n'):

l = len(t)

if l > ml: ml = l