def winsorize(df: pd.DataFrame,
by: Union[str, Iterable[str]],
p: Iterable[Union[tuple, float]] = (.01, .99)):
"""Drop variables in `by' outside quantiles `p`."""
# TODO: Some kind of warning/error if too fine of quantiles are
# requested for the number of rows, e.g. .99 with 5 rows.
df = df.copy()
by = cast(Iterable[str], force_iterable(by))
# Allow different cutoffs for different variables
if hasattr(p[0], '__iter__'):
assert len(p) == len(by)
else:
p = [p] * len(by)
survive_winsor = np.array([True] * df.shape[0])
for idx, col in enumerate(by):
cuts = df[col].quantile(p[idx]).values
survive_this = np.logical_and(df[col] >= cuts[0], df[col] <= cuts[1])
survive_winsor = np.minimum(survive_winsor, survive_this)
df = df[survive_winsor]
return df
def outreg(regs,
var_names=None,
var_labels=None,
digits=4,
stars=True,
se="(",
options=False):
"""Create the guts of a Latex tabular enviornment from regression results.
Args:
regs (Results or iterable of Results): Regressions to output to table.
var_names (str or iterable of str): Variable names to pull from regs.
If none specified, by default uses the pandas dataframe colum
names.
var_labels (str or iterable of str): Pretty names for variables in
table. If none specified, will use var_names.
Keyword Args:
digits (int): Defaults to 4. How many digits to include past decimal.
stars (bool): Defaults to True. If True, adds stars to mark statistical
significance.
se (str): Defaults to "(". Marker for standard errors. May
also choose brackets with ``se="["``.
options (bool): Default to False: If True, return a ``dict`` with
formatting options that were generated by ``outreg``:
``name_just``, ``stat_just``, etc., for additional calls to
``table_mainrow`` and ``table_statrow``.
Return:
str: LaTeX fragment meant to be wrapped in a tabular environment.
"""
regs = force_iterable(regs)
if var_names is None:
var_names = regs[0].beta.index.tolist()
if len(regs) > 1:
for reg in regs[1:]:
var_names += [
x for x in reg.beta.index.tolist() if x not in var_names
]
if var_labels is None:
var_labels = [x if type(x) is str else str(x) for x in var_names]
opt_dict = _set_options(var_labels, digits, stars)
table_str = ''
for var_idx, varname in enumerate(var_names):
table_str += table_mainrow(var_labels[var_idx], varname, regs,
**opt_dict)
if options:
return table_str, opt_dict
else:
return table_str
def table_mainrow(rowname, varname, regs,
name_just=24, stat_just=12, digits=3, se="(",
stars=True):
"""Add a table row of regression coefficients with standard errors.
Args:
rowname (str): First cell of table row, i.e., the row's name.
varname (str): Name of variable to pull from ``Results`` object.
regs (Results or iterable of Results): Regressions from which
to pull coefficients named ``varname``.
Keyword Args:
name_just (int):
stat_just (int):
digits (int):
se (str):
stars (bool):
Returns:
str: String of table row.
"""
# Start beta and SE rows
beta_vals = []
se_vals = []
# Extract beta/sig and se values to pass to `table_statrow`
for reg in force_iterable(regs):
if type(reg) is not Results or varname not in reg.beta:
beta_vals.append('')
se_vals.append('')
else:
# Beta and stars
this_beta = _format_nums(reg.beta[varname], digits=digits)
if stars:
this_sig = _sig_level(reg.pt[varname])
else:
this_sig = ''
beta_vals.append(this_beta + this_sig)
# Standard Error
this_se = reg.se[varname]
se_vals.append(this_se)
beta_row = table_statrow(rowname, beta_vals, name_just=name_just,
stat_just=stat_just)
se_row = table_statrow('', se_vals, name_just=name_just,
stat_just=stat_just, sd=se, digits=digits)
full_row = beta_row + se_row
return full_row
def table_mainrow(rowname, varname, regs,
name_just=24, stat_just=12, digits=3, se="(",
stars=True):
"""Add a table row of regression coefficients with standard errors.
Args:
rowname (str): First cell of table row, i.e., the row's name.
varname (str): Name of variable to pull from ``Results`` object.
regs (Results or iterable of Results): Regressions from which
to pull coefficients named ``varname``.
Keyword Args:
name_just (int):
stat_just (int):
digits (int):
se (str):
stars (bool):
Returns:
str: String of table row.
"""
# Start beta and SE rows
beta_vals = []
se_vals = []
# Extract beta/sig and se values to pass to `table_statrow`
for reg in force_iterable(regs):
if type(reg) is not Results or varname not in reg.beta:
beta_vals.append('')
se_vals.append('')
else:
# Beta and stars
this_beta = _format_nums(reg.beta[varname], digits=digits)
if stars:
this_sig = _sig_level(reg.pt[varname])
else:
this_sig = ''
beta_vals.append(this_beta + this_sig)
# Standard Error
this_se = reg.se[varname]
se_vals.append(this_se)
beta_row = table_statrow(rowname, beta_vals, name_just=name_just,
stat_just=stat_just)
se_row = table_statrow('', se_vals, name_just=name_just,
stat_just=stat_just, sd=se, digits=digits)
full_row = beta_row + se_row
return full_row
def outreg(regs, var_names=None, var_labels=None, digits=4, stars=True, se="(",
options=False):
"""Create the guts of a Latex tabular enviornment from regression results.
Args:
regs (Results or iterable of Results): Regressions to output to table.
var_names (str or iterable of str): Variable names to pull from regs.
If none specified, by default uses the pandas dataframe colum
names.
var_labels (str or iterable of str): Pretty names for variables in
table. If none specified, will use var_names.
Keyword Args:
digits (int): Defaults to 4. How many digits to include past decimal.
stars (bool): Defaults to True. If True, adds stars to mark statistical
significance.
se (str): Defaults to "(". Marker for standard errors. May
also choose brackets with ``se="["``.
options (bool): Default to False: If True, return a ``dict`` with
formatting options that were generated by ``outreg``:
``name_just``, ``stat_just``, etc., for additional calls to
``table_mainrow`` and ``table_statrow``.
Return:
str: LaTeX fragment meant to be wrapped in a tabular environment.
"""
regs = force_iterable(regs)
if var_names is None:
var_names = regs[0].beta.index.tolist()
if len(regs) > 1:
for reg in regs[1:]:
var_names += [x for x in reg.beta.index.tolist()
if x not in var_names]
if var_labels is None:
var_labels = [x if type(x) is str else str(x) for x in var_names]
opt_dict = _set_options(var_labels, digits, stars)
table_str = ''
for var_idx, varname in enumerate(var_names):
table_str += table_mainrow(var_labels[var_idx], varname, regs,
**opt_dict)
if options:
return table_str, opt_dict
else:
return table_str
def test_string(self):
a_string = 'abcd'
expected = (a_string,)
result = force_iterable(a_string)
assert expected == result
def test_array(self):
expected = np.arange(3)
result = force_iterable(expected)
assert_array_equal(expected, result)
def test_tup(self):
expected = (1, 2, 3)
result = force_iterable(expected)
assert expected == result
def test_int(self):
an_int = 10
expected = (an_int,)
result = force_iterable(an_int)
assert expected == result
def test_list(self):
expected = [1, 2, 3]
result = force_iterable(expected)
assert expected == result