def __init__(self, low, high):
if low is None or high is None:
raise GgplotError("Arguments to", self.__class__.__name__,
"cannot be None")
try:
_ = high - low
except TypeError:
raise GgplotError("Arguments to", self.__class__.__name__,
"must be of a numeric type")
self.low, self.high = low, high
def _calculate(self, data):
x = pop(data, 'x', None)
y = pop(data, 'y', None)
# intercept and slope may be one of:
# - aesthetics to geom_abline or
# - parameter settings to stat_abline
slope = pop(data, 'slope', self.params['slope'])
intercept = pop(data, 'intercept', self.params['intercept'])
if hasattr(slope, '__call__'):
if x is None or y is None:
raise GgplotError(
'To compute the slope, x & y aesthetics are needed')
try:
slope = slope(x, y)
except TypeError as err:
raise GgplotError(*err.args)
if hasattr(intercept, '__call__'):
if x is None or y is None:
raise GgplotError(
'To compute the intercept, x & y aesthetics are needed')
try:
intercept = intercept(x, y)
except TypeError as err:
raise GgplotError(*err.args)
try:
n = len(slope)
except TypeError:
n = 1
try:
_n = len(intercept)
except TypeError:
_n = 1
if n != _n:
raise GgplotError('Specified {} slopes but {} intercepts'.format(
n, _n))
slope = make_iterable(slope)
intercept = make_iterable(intercept)
new_data = pd.DataFrame({'slope': slope, 'intercept': intercept})
# Copy the other aesthetics into the new dataframe
n = len(slope)
for ae in data:
new_data[ae] = make_iterable_ntimes(data[ae].iloc[0], n)
return new_data
def _calculate(self, data):
x = data.pop('x')
try:
float(x.iloc[0])
except:
try:
# try to use it as a pandas.tslib.Timestamp
x = [ts.toordinal() for ts in x]
except:
raise GgplotError("stat_density(): aesthetic x mapping " +
"needs to be convertable to float!")
# TODO: Implement weight
try:
weight = data.pop('weight')
except KeyError:
weight = np.ones(len(x))
# TODO: Get "full" range of densities
# i.e tail off to zero like ggplot2? But there is nothing
# wrong with the current state.
kde = gaussian_kde(x)
bottom = np.min(x)
top = np.max(x)
step = (top - bottom) / 1000.0
x = np.arange(bottom, top, step)
y = kde.evaluate(x)
new_data = pd.DataFrame({'x': x, 'y': y})
# Copy the other aesthetics into the new dataframe
n = len(x)
for ae in data:
new_data[ae] = make_iterable_ntimes(data[ae].iloc[0], n)
return new_data
def _find_aes_and_data(self, args, kwargs):
"""
Identify the aes and data objects.
Return a dictionary of the aes mappings and
the data object.
- args is a list
- kwargs is a dictionary
Note: This is a helper function for self.__init__
It modifies the kwargs
"""
passed_aes = {}
data = None
aes_err = 'Found more than one aes argument. Expecting zero or one'
for arg in args:
if isinstance(arg, aes) and passed_aes:
raise Execption(aes_err)
if isinstance(arg, aes):
passed_aes = arg
elif isinstance(arg, pd.DataFrame):
data = arg
else:
raise GgplotError(
'Unknown argument of type "{0}".'.format(type(arg)))
if 'mapping' in kwargs and passed_aes:
raise GgplotError(aes_err)
elif not passed_aes and 'mapping' in kwargs:
passed_aes = kwargs.pop('mapping')
if data is None and 'data' in kwargs:
data = kwargs.pop('data')
_aes = {}
# To make mapping of columns to geom/stat or stat parameters
# possible
_keep = set(self.DEFAULT_PARAMS) | set(self._stat_type.DEFAULT_PARAMS)
for k, v in passed_aes.items():
if k in self.valid_aes or k in _keep:
_aes[k] = v
else:
raise GgplotError('Cannot recognize aesthetic: %s' % k)
return _aes, data, kwargs
def __init__(self, low=None, high=None):
if low != None:
try:
_ = low - 0
except TypeError:
raise GgplotError("The 'low' argument to",
self.__class__.__name__,
"must be of a numeric type or None")
if high != None:
try:
_ = high - 0
except TypeError:
raise GgplotError("The 'high' argument to",
self.__class__.__name__,
"must be of a numeric type or None")
self.low, self.high = low, high
def __init__(self, x=None, y=None, ncol=None, nrow=None, scales="free"):
if x is None and y is None:
raise GgplotError(
"You need to specify a variable name: facet_wrap('var')")
add_ggplotrc_params(self)
self.x = x
self.y = y
self.ncol = ncol
self.nrow = nrow
self.scales = scales
def _calculate_global(self, data):
# Calculate breaks if x is not categorical
binwidth = self.params['binwidth']
self.breaks = self.params['breaks']
right = self.params['right']
x = data['x'].values
# For categorical data we set labels and x-vals
if is_categorical(x):
labels = self.params['labels']
if labels == None:
labels = sorted(set(x))
self.labels = labels
self.length = len(self.labels)
# For non-categoriacal data we set breaks
if not (is_categorical(x) or self.breaks):
# Check that x is numerical
if len(x) > 0 and isinstance(x[0], datetime.date):
def convert(d):
d = datetime.datetime.combine(d,
datetime.datetime.min.time())
return time.mktime(d.timetuple())
x = [convert(d) for d in x]
elif len(x) > 0 and isinstance(x[0], datetime.datetime):
x = [time.mktime(d.timetuple()) for d in x]
elif len(x) > 0 and isinstance(x[0], datetime.time):
raise GgplotError("Cannot recognise the type of x")
elif not cbook.is_numlike(x[0]):
raise GgplotError("Cannot recognise the type of x")
if binwidth is None:
_bin_count = 30
self._print_warning(_MSG_BINWIDTH)
else:
_bin_count = int(np.ceil(np.ptp(x))) / binwidth
_, self.breaks = pd.cut(x,
bins=_bin_count,
labels=False,
right=right,
retbins=True)
self.length = len(self.breaks)
def _verify_aesthetics(self, data):
"""
Check if all the required aesthetics have been specified
Raise an Exception if an aesthetic is missing
"""
missing_aes = self.REQUIRED_AES - set(data.columns)
if missing_aes:
msg = '{} requires the following missing aesthetics: {}'
raise GgplotError(
msg.format(self.__class__.__name__, ', '.join(missing_aes)))
def __radd__(self, gg):
x = gg.data.get(self.x)
y = gg.data.get(self.y)
if x is None and y is None:
raise GgplotError("No facets provided!")
# only do the deepcopy after the check
gg = deepcopy(gg)
if x is None:
n_dim_x = 1
else:
n_dim_x = x.nunique()
if y is None:
n_dim_y = 1
else:
n_dim_y = y.nunique()
n_dim = n_dim_x * n_dim_y
if self.ncol is None and self.nrow is None:
n_rows = n_dim_x
n_cols = n_dim_y
elif self.nrow is None:
n_rows = self.ncol
n_cols = math.ceil(float(n_dim) / n_rows)
elif self.ncol is None:
n_cols = self.nrow
n_rows = math.ceil(float(n_dim) / n_cols)
else:
n_rows = self.ncol
n_cols = self.nrow
gg.n_rows, gg.n_columns = int(n_rows), int(n_cols)
facets = []
if self.x:
facets.append(self.x)
if self.y:
facets.append(self.y)
gg.facets = facets
gg.facet_type = "grid"
gg.facet_scales = self.scales
combos = []
for x_i in sorted(x.unique()):
if y is not None:
for y_i in sorted(y.unique()):
combos.append((x_i, y_i))
else:
combos.append((x_i, 1))
gg.facet_pairs = combos
return gg
def __init__(self, *args, **kwargs):
self.valid_aes = set(self.DEFAULT_AES) ^ self.REQUIRED_AES
self._stat_type = self._get_stat_type(kwargs)
self.aes, self.data, kwargs = self._find_aes_and_data(args, kwargs)
# This set will list the geoms that were uniquely set in this
# geom (not specified already i.e. in the ggplot aes).
self.aes_unique_to_geom = set(self.aes.keys())
if 'colour' in kwargs:
kwargs['color'] = kwargs.pop('colour')
# When a geom is created, some of the parameters may be meant
# for the stat and some for the layer.
# Some arguments are can be identified as either aesthetics to
# the geom and or parameter settings to the stat, in this case
# if the argument has a scalar value it is a setting for the stat.
self._stat_params = {}
self.params = deepcopy(self.DEFAULT_PARAMS)
self.manual_aes = {}
for k, v in kwargs.items():
if k in self.aes:
raise GgplotError('Aesthetic, %s, specified twice' % k)
elif (k in self.valid_aes and k in self._stat_type.DEFAULT_PARAMS
and is_scalar_or_string(kwargs[k])):
self._stat_params[k] = v
elif k in self.valid_aes:
self.manual_aes[k] = v
elif k in self.DEFAULT_PARAMS:
self.params[k] = v
elif k in self._stat_type.DEFAULT_PARAMS:
self._stat_params[k] = v
else:
raise GgplotError('Cannot recognize argument: %s' % k)
self._cache = {}
# When putting together the plot information for the geoms,
# we need the aethetics names to be matplotlib compatible.
# These are created and stored in self._cache and so would
# go stale if users or geoms change geom.manual_aes
self._create_aes_with_mpl_names()
def _calculate(self, data):
x = pop(data, 'x', None)
# xintercept may be one of:
# - aesthetic to geom_vline or
# - parameter setting to stat_vline
xintercept = pop(data, 'xintercept', self.params['xintercept'])
if hasattr(xintercept, '__call__'):
if x is None:
raise GgplotError(
'To compute the intercept, x aesthetic is needed')
try:
xintercept = xintercept(x)
except TypeError as err:
raise GgplotError(*err.args)
xintercept = make_iterable(xintercept)
new_data = pd.DataFrame({'xintercept': xintercept})
# Copy the other aesthetics into the new dataframe
n = len(xintercept)
for ae in data:
new_data[ae] = make_iterable_ntimes(data[ae].iloc[0], n)
return new_data
def __init__(self, title):
if title is None:
raise GgplotError("Arguments to", self.__class__.__name__,
"cannot be None")
self.title = title
def __init__(self, ylab):
if ylab is None:
raise GgplotError("Arguments to", self.__class__.__name__,
"cannot be None")
self.ylab = ylab
def _calculate(self, data):
x = data.pop('x')
right = self.params['right']
# y values are not needed
try:
del data['y']
except KeyError:
pass
else:
self._print_warning(_MSG_YVALUE)
if len(x) > 0 and isinstance(x.get(0), datetime.date):
def convert(d):
d = datetime.datetime.combine(d, datetime.datetime.min.time())
return time.mktime(d.timetuple())
x = x.apply(convert)
elif len(x) > 0 and isinstance(x.get(0), datetime.datetime):
x = x.apply(lambda d: time.mktime(d.timetuple()))
elif len(x) > 0 and isinstance(x.get(0), datetime.time):
raise GgplotError("Cannot recognise the type of x")
# If weight not mapped to, use one (no weight)
try:
weights = data.pop('weight')
except KeyError:
weights = np.ones(len(x))
else:
weights = make_iterable_ntimes(weights, len(x))
if is_categorical(x.values):
x_assignments = x
x = self.labels
width = make_iterable_ntimes(self.params['width'], self.length)
elif cbook.is_numlike(x.iloc[0]):
x_assignments = pd.cut(x, bins=self.breaks, labels=False,
right=right)
width = np.diff(self.breaks)
x = [self.breaks[i] + width[i] / 2
for i in range(len(self.breaks)-1)]
else:
raise GgplotError("Cannot recognise the type of x")
# Create a dataframe with two columns:
# - the bins to which each x is assigned
# - the weights of each x value
# Then create a weighted frequency table
_df = pd.DataFrame({'assignments': x_assignments,
'weights': weights
})
_wfreq_table = pd.pivot_table(_df, values='weights',
rows=['assignments'], aggfunc=np.sum)
# For numerical x values, empty bins get have no value
# in the computed frequency table. We need to add the zeros and
# since frequency table is a Series object, we need to keep it ordered
try:
empty_bins = set(self.labels) - set(x_assignments)
except:
empty_bins = set(range(len(width))) - set(x_assignments)
_wfreq_table = _wfreq_table.to_dict()
for _b in empty_bins:
_wfreq_table[_b] = 0
_wfreq_table = pd.Series(_wfreq_table).sort_index()
y = list(_wfreq_table)
new_data = pd.DataFrame({'x': x, 'y': y, 'width': width})
# Copy the other aesthetics into the new dataframe
n = len(x)
for ae in data:
new_data[ae] = make_iterable_ntimes(data[ae].iloc[0], n)
return new_data
def _calculate(self, data):
x = data.pop('x')
breaks = self.params['breaks']
right = self.params['right']
binwidth = self.params['binwidth']
# y values are not needed
try:
del data['y']
except KeyError:
pass
else:
self._print_warning(_MSG_YVALUE)
# If weight not mapped to, use one (no weight)
try:
weights = data.pop('weight')
except KeyError:
weights = np.ones(len(x))
else:
weights = make_iterable_ntimes(weights, len(x))
categorical = is_categorical(x.values)
if categorical:
x_assignments = x
x = sorted(set(x))
width = make_iterable_ntimes(self.params['width'], len(x))
elif cbook.is_numlike(x.iloc[0]):
if breaks is None and binwidth is None:
_bin_count = 30
self._print_warning(_MSG_BINWIDTH)
if binwidth:
_bin_count = int(np.ceil(np.ptp(x))) / binwidth
# Breaks have a higher precedence and,
# pandas accepts either the breaks or the number of bins
_bins_info = breaks or _bin_count
x_assignments, breaks = pd.cut(x, bins=_bins_info, labels=False,
right=right, retbins=True)
width = np.diff(breaks)
x = [breaks[i] + width[i] / 2
for i in range(len(breaks)-1)]
else:
raise GgplotError("Cannot recognise the type of x")
# Create a dataframe with two columns:
# - the bins to which each x is assigned
# - the weights of each x value
# Then create a weighted frequency table
_df = pd.DataFrame({'assignments': x_assignments,
'weights': weights
})
_wfreq_table = pd.pivot_table(_df, values='weights',
rows=['assignments'], aggfunc=np.sum)
# For numerical x values, empty bins get have no value
# in the computed frequency table. We need to add the zeros and
# since frequency table is a Series object, we need to keep it ordered
if len(_wfreq_table) < len(x):
empty_bins = set(range(len(x))) - set(x_assignments)
for _b in empty_bins:
_wfreq_table[_b] = 0
_wfreq_table = _wfreq_table.sort_index()
y = list(_wfreq_table)
new_data = pd.DataFrame({'x': x, 'y': y, 'width': width})
# Copy the other aesthetics into the new dataframe
n = len(x)
for ae in data:
new_data[ae] = make_iterable_ntimes(data[ae].iloc[0], n)
return new_data