def groupby(df: "CERO",
*args,
key: "Union[int, list[int]]" = None,
match: str = None,
agg: str = None,
**kwargs):
if key is None:
raise TypeError(
"'key' must be provided to 'groupby' function as either an int or list of ints."
)
elif not issubclass(type(key), list):
key = [key]
if not all([issubclass(type(k), int) for k in key]):
raise TypeError(
"'key' must be provided to 'groupby' function as either an int or list of ints."
)
defaults = {"axis": 0, "sort": False, "group_keys": False}
defaults.update(kwargs)
match = _Identifier.tupleize_name(match)
m_ids = [match]
if match is None:
m_ids = _Identifier.unique_id_fields(df.index.values, key=key)
conv = lambda x: tuple(x) if issubclass(type(x), str) else x
m_ids = [conv(m) for m in m_ids]
rename_dict = {}
for m in m_ids:
# Create func that identifies rows for grouping
def f(x):
return all([x[k] == m[idx] for idx, k in enumerate(key)])
# Groupby and apply aggregation function
agg_df = df.groupby(by=f, **defaults).agg(agg)
# Put aggregated calculation in first row that meets the condition
row_loc = next(x for x in df.index.values if f(x))
df.iloc[df.index.get_loc(row_loc)] = agg_df.loc[True]
# Rename row
rename_dict.update(
{row_loc: _Identifier.keep_only_fields(key, row_loc)[0]})
CERO.rename_index_values(df, rename_dict, inplace=True)
return df
def _set_inputs(self, cero: pd.DataFrame):
"""Copies each data series in ``cero`` indexed by the items in ``inp_list`` to an ``OrderedDict``. This \
ensures that ``operations`` do not alter ``cero``.
"""
if self["inputs"] == []:
# Input is entire CERO unless otherwise specified
self["inputs"] = cero.index.tolist()
# Check values in dataframe - check is necessary because KeyError is not thrown if some values are in index (pandas version 0.22).
invalid_inputs = [i for i in self["inputs"] if i not in cero.index]
try:
self.inputs = copy.deepcopy(cero.iloc[[cero.index.get_loc(loc) for loc in self["inputs"]]]) # Reduce data frame to necessary data and copy
except KeyError:
invalid_inputs = [i for i in self["inputs"] if i not in cero.index]
msg = ("Inputs %s do not exist. The most likely reason is that the configuration file is " +
"incorrectly specified, or lacks specification. If debugging level has been set to " +
"'DEBUG', then the input list is in the log file - note that this list may be " +
"extraordinarily long. Common causes of this problem include: \n" +
" 1. Set definition in configuration file includes elements that do not exist in the CERO.\n" +
" 2. Mis-spellings of identifiers in the configuration file (which includes names of sets for 'inputs' or 'arrays').\n" +
" 3. Incorrect ordering of sets in the identifier."
) % invalid_inputs
FromCERO._logger.error(msg)
raise KeyError(msg)
assert (isinstance(self.inputs, pd.DataFrame))
map_dict = {}
for map_op in self.get("map", []):
idx = map_op.get("idx")
orig_s = self["sets"][map_op["orig"]]
ns = self["sets"][map_op["new"]]
for val in self.inputs.index.values:
new_val = val
if idx is not None and (val[idx] in orig_s) and (not isinstance(val, str)):
new_val = val[:idx] + (ns[orig_s.index(val[idx])],) + val[idx+1:]
elif val in orig_s:
new_val = ns[orig_s.index(val)]
map_dict.update({val: new_val})
CERO.rename_index_values(self.inputs, map_dict, inplace=True)
def exec_procedures(self, cero):
""" Execute all the procedures of the FromCERO object
.
:param pandas.DataFrame cero: A CERO to serve as input for the procedures. The argument is not mutated/modified.
"""
CERO.is_cero(cero, raise_exception=True, empty_ok=False)
CERO.rename_index_values(cero, self.get("map", {}))
self.output_procedures = OrderedDict()
for procedure in self["procedures"]:
try:
ret = procedure.exec_ops(cero)
# if ret is not None, should be dict with key: procedure["name"], value: resultant CERO
except Exception as e:
raise e.__class__(e.__str__() + " Error in procedure '%s'." % (procedure["name"]))
if ret is None:
ret = {}
self.output_procedures.update(ret)
else:
if not self["procedures"]: # If empty list
self.output_procedures["default_output"] = cero
if any([not procedure.get("file") for procedure in self["procedures"]]):
msg = "It has been detected that not all procedures direct output to file. Therefore some output will go to \'%s\'." % self["file"]
print(msg)
FromCERO._logger.info(msg)
if self.output_procedures != {}:
file_ext = os.path.splitext(self["file"])[1][1:]
if file_ext in FromCERO.sup_output_types:
out_df = CERO.combine_ceros(list(self.output_procedures.values()))
FromCERO.dataframe_out(out_df, self["file"], output_type=file_ext)
elif file_ext in FromCERO._Procedure.sup_output_types:
raise ValueError("This data type is not supported for general export, because it probably has a more than 2 dimensions - export using 'procedures' instead.")
else:
raise ValueError("Unsupported data type detected for general export.")
def _exec_op(self, op: dict):
# Apply operation to procedure
func_name = op.pop('func', "noop") # Perform noop (no-operation) if no func provided.
op_args = op.pop('args', [])
rename = op.pop("rename", None)
arrays = None
if "arrays" in op:
arrays = op.pop("arrays")
if issubclass(type(arrays), str):
arrays = [arrays]
arrays = _Identifier.get_all_idents(arrays, sets=self["sets"])
for mod in self["libfuncs"]:
if hasattr(mod, func_name):
func = getattr(mod, func_name)
break
else:
msg = ('Invalid function name provided - \'%s\'. Function does not exist in any of the modules %s. It may be necessary to create a python module with the necessary functions and provide this file with the \'libfuncs\' option.' %
(func_name, self["libfuncs"]))
FromCERO._logger.error(msg)
raise AttributeError(msg)
FromCERO._logger.debug("Function call: %s(*arrays, **op)" % func.__name__)
ret = func(self.inputs, *op_args, locs=arrays, **op)
op['func'] = func.__name__ # For cleanliness of presentation
if rename is not None:
if ret is None:
ret = getattr(libfuncs, "noop")(self.inputs, *op_args, locs=arrays, **op)
if isinstance(rename, str):
rename = {ret.index.tolist()[0]: rename} # Rename the first index by default
if issubclass(type(rename), list):
# Build mapping dictionary
rename = _Identifier.get_mapping_dict(ret.index.tolist(), rename, sets=self.get("sets"))
elif issubclass(type(rename), dict):
rename = _Identifier.get_one_to_one_mapping(rename, sets=self.get("sets"))
# At this point, rename should be one-to-one mapping dict
renamed = CERO.rename_index_values(ret.loc[list(rename.keys())], rename, inplace=False)
ret = renamed.loc[list(rename.values())] # Restrict renamed to only the rows that have been specified
# Note that ret will be restricted to only those values that have been renamed.
return ret
def test_name_map(self):
def init_df():
df = pd.DataFrame.from_dict(
{
"A": [1, 2, 3],
"B": [3, 4, 5],
"C": [6, 7, 8]
},
orient="index")
df.columns = pd.DatetimeIndex(
pd.to_datetime([2017, 2018, 2019], format="%Y"))
df.sort_index(inplace=True)
df = df.astype(pd.np.float32)
self.assertTrue(CERO.is_cero(df))
return df
df = init_df()
mapping = dict([("A", "D"), ("B", "E"), ("C", "F")])
res = CERO.rename_index_values(df, mapping)
test_names = ["D", "E", "F"]
self.assertIsNone(res)
self.assertTrue(
all([x == y for (x, y) in zip(df.index.tolist(), test_names)]))
# Test 2
df = init_df()
mapping = dict([("B", "E"), ("C", "F")])
res = CERO.rename_index_values(df, mapping)
test_names = ["A", "E", "F"]
self.assertIsNone(res)
self.assertTrue(
all([x == y for (x, y) in zip(df.index.tolist(), test_names)]))
# Test 3
df = init_df()
mapping = dict([("A", "D"), ("B", "E"), ("C", "F")])
res = CERO.rename_index_values(df, mapping, inplace=False)
test_names = ["D", "E", "F"]
test_names_df = ["A", "B", "C"]
self.assertTrue(
all([x == y for (x, y) in zip(res.index.tolist(), test_names)]))
self.assertTrue(
all([x == y for (x, y) in zip(df.index.tolist(), test_names_df)]))
# Test 4
df = init_df()
mapping = dict([("B", "E"), ("C", "F")])
res = CERO.rename_index_values(df, mapping, inplace=False)
test_names = ["A", "E", "F"]
test_names_df = ["A", "B", "C"]
self.assertTrue(
all([x == y for (x, y) in zip(res.index.tolist(), test_names)]))
self.assertTrue(
all([x == y for (x, y) in zip(df.index.tolist(), test_names_df)]))
