def cast_any(self, value, fmt=None):
if isinstance(value, self.python_type):
return value
try:
return parse_datetime_day_first(value)
except (TypeError, ValueError) as e:
raise_with_traceback(InvalidDateType(e))
def cast_any(self, value, fmt=None):
if isinstance(value, self.python_type):
return value
try:
return date_parse(value, dayfirst=True).date()
except (TypeError, ValueError) as e:
raise_with_traceback(InvalidDateType(e))
def inner(*args, **kwargs):
"""Wraps specified exceptions"""
try:
return a_func(*args, **kwargs)
# pylint: disable=catching-non-exception
except tuple(errors) as exception:
raise_with_traceback(GaxError('RPC failed', cause=exception))
def _get_channel_number(section_name):
match = re.match(r'^.*_channel<(\d{4})>$', section_name)
if match:
result = int(match.group(1))
return result
else:
raise_with_traceback(RuntimeError("Unable to detect channel name from \"%s\""%section_name))
def _type_check(self, type_constraint, datum, is_input):
"""Typecheck a PTransform related datum according to a type constraint.
This function is used to optionally type-check either an input or an output
to a PTransform.
Args:
type_constraint: An instance of a typehints.TypeContraint, one of the
white-listed builtin Python types, or a custom user class.
datum: An instance of a Python object.
is_input: True if 'datum' is an input to a PTransform's DoFn. False
otherwise.
Raises:
TypeError: If 'datum' fails to type-check according to 'type_constraint'.
"""
datum_type = 'input' if is_input else 'output'
try:
check_constraint(type_constraint, datum)
except CompositeTypeHintError as e:
raise_with_traceback(TypeCheckError(e.args[0]))
except SimpleTypeHintError:
error_msg = ("According to type-hint expected %s should be of type %s. "
"Instead, received '%s', an instance of type %s."
% (datum_type, type_constraint, datum, type(datum)))
raise_with_traceback(TypeCheckError(error_msg))
def _run_pants_with_retry(self, port, retries=3):
"""Runs pants remotely with retry and recovery for nascent executions."""
attempt = 1
while 1:
logger.debug(
'connecting to pantsd on port {} (attempt {}/{})'.format(port, attempt, retries)
)
try:
return self._connect_and_execute(port)
except self.RECOVERABLE_EXCEPTIONS as e:
if attempt > retries:
raise self.Fallback(e)
self._backoff(attempt)
logger.warn(
'pantsd was unresponsive on port {}, retrying ({}/{})'
.format(port, attempt, retries)
)
# One possible cause of the daemon being non-responsive during an attempt might be if a
# another lifecycle operation is happening concurrently (incl teardown). To account for
# this, we won't begin attempting restarts until at least 1 second has passed (1 attempt).
if attempt > 1:
port = self._restart_pantsd()
attempt += 1
except NailgunClient.NailgunError as e:
# Ensure a newline.
logger.fatal('')
logger.fatal('lost active connection to pantsd!')
raise_with_traceback(
self.Terminated,
'abruptly lost active connection to pantsd runner: {!r}'.format(e)
)
def connect(self):
"""
This method causes the `Duct` instance to connect to the service, if it
is not already connected. It is not normally necessary for a user to
manually call this function, since when a connection is required, it is
automatically created.
Subclasses should implement `Duct._connect` to do whatever is necessary
to bring a connection into being.
Compared to base `Duct.connect`, this method will automatically catch
the first `DuctAuthenticationError` error triggered by `Duct.connect`
if any smartcards have been configured, before trying once more.
Returns:
`Duct` instance: A reference to the current object.
"""
try:
Duct.connect(self)
except DuctServerUnreachable as e:
raise_with_traceback(e)
except DuctAuthenticationError as e:
if self.smartcards and self.prepare_smartcards():
Duct.connect(self)
else:
raise_with_traceback(e)
return self
def jsonify(obj, pretty=False):
"""
Turn a nested object into a (compressed) JSON string.
Parameters
----------
obj : dict
Any kind of dictionary structure.
pretty : bool, optional
Whether to format the resulting JSON in a more legible way (
default False).
"""
if pretty:
params = dict(sort_keys=True, indent=2, allow_nan=False,
separators=(",", ": "), ensure_ascii=False)
else:
params = dict(sort_keys=False, indent=None, allow_nan=False,
separators=(",", ":"), ensure_ascii=False)
try:
return json.dumps(obj, **params)
except (TypeError, ValueError) as error:
LOGGER.critical(
"The memote result structure is incompatible with the JSON "
"standard.")
log_json_incompatible_types(obj)
raise_with_traceback(error)
def __call__(self, *args, **kwargs):
if not current_app.config.get('INDEXING_ENABLED', True):
return self._route(*args, **kwargs)
log = PageView(
page=request.full_path,
endpoint=request.endpoint,
user_id=current_user.id,
ip_address=request.remote_addr,
version=__version__
)
errorlog = None
log.object_id, log.object_type, log.object_action, reextract_after_request = self.extract_objects(*args, **kwargs)
db_session.add(log) # Add log here to ensure pageviews are accurate
try:
return self._route(*args, **kwargs)
except Exception as e:
db_session.rollback() # Ensure no lingering database changes remain after crashed route
db_session.add(log)
errorlog = ErrorLog.from_exception(e)
db_session.add(errorlog)
db_session.commit()
raise_with_traceback(e)
finally:
# Extract object id and type after response generated (if requested) to ensure
# most recent data is collected
if reextract_after_request:
log.object_id, log.object_type, log.object_action, _ = self.extract_objects(*args, **kwargs)
if errorlog is not None:
log.id_errorlog = errorlog.id
db_session.add(log)
db_session.commit()
def inner(*args, **kwargs):
"""Wraps specified exceptions"""
try:
return a_func(*args, **kwargs)
# pylint: disable=catching-non-exception
except tuple(to_catch) as exception:
utils.raise_with_traceback(errors.create_error("RPC failed", cause=exception))
def start_response(status, response_headers, exc_info=None):
if exc_info:
try:
if headers_sent:
# Re-raise if too late
raise_with_traceback(exc_info[0](exc_info[1]))
finally:
exc_info = None # avoid dangling circular ref
else:
assert not headers_set, 'Headers already set!'
assert type(status) is str, 'Status must be a string'
assert len(status) >= 4, 'Status must be at least 4 characters'
assert int(status[:3]), 'Status must begin with 3-digit code'
assert status[3] == ' ', 'Status must have a space after code'
assert type(response_headers) is list, 'Headers must be a list'
if FCGI_DEBUG:
logging.debug('response headers:')
for name, val in response_headers:
assert type(name) is str, 'Header name "%s" must be a string' % name
assert type(val) is str, 'Value of header "%s" must be a string' % name
logging.debug('%s: %s' % (name, val))
headers_set[:] = [status, response_headers]
return write
def cast_default(self, value):
try:
struct_time = time.strptime(value, self.ISO8601)
return datetime.time(struct_time.tm_hour, struct_time.tm_min,
struct_time.tm_sec)
except (TypeError, ValueError) as e:
raise_with_traceback(exceptions.InvalidTimeType(e))
def _execute(self, statement, cursor, wait, session_properties):
"""
If something goes wrong, `PrestoClient` will attempt to parse the error
log and present the user with useful debugging information. If that fails,
the full traceback will be raised instead.
"""
from pyhive import presto # Imported here due to slow import performance in Python 3
from pyhive.exc import DatabaseError # Imported here due to slow import performance in Python 3
try:
cursor = cursor or presto.Cursor(
host=self.host, port=self.port, username=self.username, password=self.password,
catalog=self.catalog, schema=self.schema, session_props=session_properties,
poll_interval=1, source=self.source, protocol=self.server_protocol
)
cursor.execute(statement)
status = cursor.poll()
if wait:
logger.progress(0)
# status None means command executed successfully
# See https://github.com/dropbox/PyHive/blob/master/pyhive/presto.py#L234
while status is not None and status['stats']['state'] != "FINISHED":
if status['stats'].get('totalSplits', 0) > 0:
pct_complete = round(status['stats']['completedSplits'] / float(status['stats']['totalSplits']), 4)
logger.progress(pct_complete * 100)
status = cursor.poll()
logger.progress(100, complete=True)
return cursor
except (DatabaseError, pandas.io.sql.DatabaseError) as e:
# Attempt to parse database error, before ultimately reraising the same
# exception, maintaining the full stacktrace.
exception, exception_args, traceback = sys.exc_info()
try:
message = e.args[0]
if isinstance(message, six.string_types):
message = ast.literal_eval(re.match("[^{]*({.*})[^}]*$", message).group(1))
linenumber = message['errorLocation']['lineNumber'] - 1
splt = statement.splitlines()
splt[linenumber] += ' <-- {errorType} ({errorName}) occurred. {message} '.format(**message)
context = '\n\n[Error Context]\n{}\n'.format('\n'.join([splt[l] for l in range(max(linenumber - 1, 0),
min(linenumber + 2, len(splt)))]))
class ErrContext(object):
def __repr__(self):
return context
# logged twice so that both notebook and console users see the error context
exception_args.args = [exception_args, ErrContext()]
logger.error(context)
except:
logger.warn(("Omniduct was unable to parse the database error messages. Refer to the "
"traceback below for full error details."))
if isinstance(exception, type):
exception = exception(exception_args)
raise_with_traceback(exception, traceback)
def __setattr__(self, name, value):
if name not in self._parameters.keys():
return object.__setattr__(self, name, value)
else:
if self._parameters[name][1] == type(value):
self._parameters[name] = (value, type(value))
else:
raise_with_traceback(TypeError("Invalid type %s for parameter \"%s\""%(type(value), name)))
def reraise():
"""Reraise the current contextmanager exception, if any"""
typ,val,tb = _exc_info.get(get_ident(), nones)
if typ:
try:
raise_with_traceback(typ(val))
finally:
del typ,val,tb
def wrapped(*args, **kwargs):
try:
return function(*args, **kwargs)
except Exception as e:
db_session.rollback()
db_session.add(ErrorLog.from_exception(e))
db_session.commit()
raise_with_traceback(e)
def raise_critical_error(self, err):
"""
This logs the error, releases any lock files and throws an exception.
The expectation is that the application exits after this.
"""
self.logger.critical(err)
self._run_exit_hooks()
raise_with_traceback(CriticalApplicationError(err))
def exception_traceback_example():
""" Throw an exception with traceback
>>> exception_traceback_example()
Traceback (most recent call last):
ValueError: exceptional
"""
from future.utils import raise_with_traceback
raise_with_traceback(ValueError('exceptional'))
def wrapper(self, method, args, kwargs):
try:
result = method(*args, **kwargs)
except TypeCheckError as e:
error_msg = ('Runtime type violation detected within ParDo(%s): '
'%s' % (self.full_label, e))
raise_with_traceback(TypeCheckError(error_msg))
else:
return self._check_type(result)
def cast_binary(self, value):
if not self._type_check(value):
raise exceptions.InvalidStringType()
try:
base64.b64decode(value)
except binascii.Error as e:
raise_with_traceback(exceptions.InvalidBinary(e))
return value
def _raise_deferred_exc(self):
"""Raises deferred exceptions from the daemon's synchronous path in the post-fork client."""
if self._deferred_exception:
try:
exc_type, exc_value, exc_traceback = self._deferred_exception
raise_with_traceback(exc_value, exc_traceback)
except TypeError:
# If `_deferred_exception` isn't a 3-item tuple (raising a TypeError on the above
# destructuring), treat it like a bare exception.
raise self._deferred_exception
def cast_default(self, value):
if isinstance(value, self.py):
return value
try:
json_value = json.loads(value)
if isinstance(json_value, self.py):
return json_value
else:
raise exceptions.InvalidObjectType()
except (TypeError, ValueError) as e:
raise_with_traceback(exceptions.InvalidObjectType(e))
def decode(data, encoding='utf-8'):
assert encoding == 'utf-8', "Only UTF-8 encoding is currently supported."
if encoding is not None:
try:
data = data.decode(encoding)
except Exception as e:
if os.environ.get('DEBUG'):
raise_with_traceback(e)
logger.warning("An decoding error has occurred... continuing anyway. To capture these errors, rerun the current command prefixed with `DEBUG=1 `.")
data = data.decode(encoding, errors='ignore')
return data
def logging_scope(func, *args, **kwargs):
logger._scope_enter(name, *wargs, **wkwargs)
success = True
try:
f = func(*args, **kwargs)
return f
except Exception as e:
success = False
raise_with_traceback(e)
finally:
logger._scope_exit(success)
def _raise_deferred_exc(self):
"""Raises deferred exceptions from the daemon's synchronous path in the post-fork client."""
if self._deferred_exception:
try:
# Expect `_deferred_exception` to be a 3-item tuple of the values returned by sys.exc_info().
# This permits use the 3-arg form of the `raise` statement to preserve the original traceback.
exc_type, exc_value, exc_traceback = self._deferred_exception
raise_with_traceback(exc_type(exc_value), exc_traceback)
except ValueError:
# If `_deferred_exception` isn't a 3-item tuple, treat it like a bare exception.
raise self._deferred_exception
def add_input(self, accumulator, element, *args, **kwargs):
if self._input_type_hint:
try:
_check_instance_type(
self._input_type_hint[0][0].tuple_types[1], element, 'element',
True)
except TypeCheckError as e:
error_msg = ('Runtime type violation detected within %s: '
'%s' % (self._label, e))
raise_with_traceback(TypeCheckError(error_msg))
return self._combinefn.add_input(accumulator, element, *args, **kwargs)
def extract_output(self, accumulator, *args, **kwargs):
result = self._combinefn.extract_output(accumulator, *args, **kwargs)
if self._output_type_hint:
try:
_check_instance_type(
self._output_type_hint.tuple_types[1], result, None, True)
except TypeCheckError as e:
error_msg = ('Runtime type violation detected within %s: '
'%s' % (self._label, e))
raise_with_traceback(TypeCheckError(error_msg))
return result
def _get_attr(self, attr_name, default_value=None):
try:
attr_value = self.attrs[attr_name]
except KeyError:
if default_value is None:
raise_with_traceback(IpcMessageException("Missing attribute " + attr_name))
else:
attr_value = default_value
return attr_value
def cast_uuid(self, value):
"""Return `value` if is a uuid, else return False."""
if not self._type_check(value):
raise exceptions.InvalidStringType(
'{0} is not of type {1}'.format(value, self.py)
)
try:
uuid.UUID(value, version=4)
return value
except ValueError as e:
raise_with_traceback(exceptions.InvalidUUID(e))
def get_param(self, param_name, default_value=None):
try:
param_value = self.attrs['params'][param_name]
except KeyError:
if default_value is None:
raise_with_traceback(IpcMessageException("Missing parameter " + param_name))
else:
param_value = default_value
return param_value
def two_dim_dict_param(obj, param_name, key_type=string_types, value_type=None):
if not isinstance(obj, dict):
raise_with_traceback(_param_type_mismatch_exception(obj, dict, param_name))
return _check_two_dim_key_value_types(obj, key_type, param_name, value_type)
def urlencode(query, doseq=False, safe='', encoding=None, errors=None):
"""Encode a sequence of two-element tuples or dictionary into a URL query string.
If any values in the query arg are sequences and doseq is true, each
sequence element is converted to a separate parameter.
If the query arg is a sequence of two-element tuples, the order of the
parameters in the output will match the order of parameters in the
input.
The query arg may be either a string or a bytes type. When query arg is a
string, the safe, encoding and error parameters are sent the quote_plus for
encoding.
"""
if hasattr(query, "items"):
query = query.items()
else:
# It's a bother at times that strings and string-like objects are
# sequences.
try:
# non-sequence items should not work with len()
# non-empty strings will fail this
if len(query) and not isinstance(query[0], tuple):
raise TypeError
# Zero-length sequences of all types will get here and succeed,
# but that's a minor nit. Since the original implementation
# allowed empty dicts that type of behavior probably should be
# preserved for consistency
except TypeError:
ty, va, tb = sys.exc_info()
raise_with_traceback(
TypeError("not a valid non-string sequence "
"or mapping object"), tb)
l = []
if not doseq:
for k, v in query:
if isinstance(k, bytes):
k = quote_plus(k, safe)
else:
k = quote_plus(str(k), safe, encoding, errors)
if isinstance(v, bytes):
v = quote_plus(v, safe)
else:
v = quote_plus(str(v), safe, encoding, errors)
l.append(k + '=' + v)
else:
for k, v in query:
if isinstance(k, bytes):
k = quote_plus(k, safe)
else:
k = quote_plus(str(k), safe, encoding, errors)
if isinstance(v, bytes):
v = quote_plus(v, safe)
l.append(k + '=' + v)
elif isinstance(v, str):
v = quote_plus(v, safe, encoding, errors)
l.append(k + '=' + v)
else:
try:
# Is this a sufficient test.py for sequence-ness?
x = len(v)
except TypeError:
# not a sequence
v = quote_plus(str(v), safe, encoding, errors)
l.append(k + '=' + v)
else:
# loop over the sequence
for elt in v:
if isinstance(elt, bytes):
elt = quote_plus(elt, safe)
else:
elt = quote_plus(str(elt), safe, encoding, errors)
l.append(k + '=' + elt)
return str('&').join(l)
def numeric_param(obj: Any, param_name: str) -> Numeric:
if not isinstance(obj, (int, float)):
raise_with_traceback(
_param_type_mismatch_exception(obj, (int, float), param_name))
return obj
def failed(desc: str) -> NoReturn: # type: ignore[misc]
if not isinstance(desc, str):
raise_with_traceback(CheckError("desc argument must be a string"))
raise_with_traceback(
CheckError("Failure condition: {desc}".format(desc=desc)))
def subclass(obj, superclass, desc=None):
if not issubclass(obj, superclass):
raise_with_traceback(_type_mismatch_error(obj, superclass, desc))
return obj
def _score_population(predictions_location, cf_dir_location):
"""
Scores estimations of treatment effect size over the population.
Args:
predictions_location (str): Path to a single tabular file where the effect estimations are located.
Files must of tabular format
* containing 4 columns: HEADER_POP_IDX, HEADER_EFFECT_SIZE, HEADER_CI_LEFT,
HEADER_CI_RIGHT.
* delimited by TABULAR_DELIMITER.
* have FILENAME_EXTENSION extension to them.
These global variables specified above can be changed when importing the module.
cf_dir_location (str): Path to a directory containing the counter-factual files (i.e. labeled, ground-truth
data).
Files must be of tabular format
* containing 3 columns: HEADER_IND_IDX, HEADER_Y1, HEADER_Y0.
* delimited by TABULAR_DELIMITER.
* have the suffix specified in COUNTERFACTUAL_FILE_SUFFIX.
* have FILENAME_EXTENSION extension to them.
These global variables specified above can be changed when importing the module.
Returns:
pd.Series: Scores. Where Series' Index is the metric name and the value is the evaluation of that metric.
"""
ufids = os.listdir(cf_dir_location)
ufids = [
f.rsplit(COUNTERFACTUAL_FILE_SUFFIX + FILENAME_EXTENSION)[0]
for f in ufids
if f.lower().endswith(COUNTERFACTUAL_FILE_SUFFIX + FILENAME_EXTENSION)
]
# Gather scoring statistics:
ratio = pd.Series(index=ufids, name="population_ratio")
bias = pd.Series(index=ufids, name="population_bias")
ci_size = pd.Series(index=ufids, name="population_ci-size")
coverage = pd.Series(data=False,
index=ufids,
dtype=np.dtype(bool),
name="population_coverage")
# Get data: # HEADER_POP_IDX | HEADER_EFFECT_SIZE | HEADER_CI_LEFT | HEADER_CI_RIGHT
estimates = pd.read_csv(predictions_location,
index_col=HEADER_POP_IDX,
sep=TABULAR_DELIMITER)
if set(ufids) - set(estimates.index):
raise_with_traceback(
AssertionError(
"Seems there are ground-truth files with no corresponding predictions\n"
"Unmatched files are:\n" + "\n".join(
[str(i) for i in (set(ufids) - set(estimates.index))])))
true_effects = pd.Series(index=ufids)
dataset_sizes = pd.Series(index=ufids, name="size")
for ufid in ufids:
# Get the true effect:
gt = pd.read_csv(os.path.join(
cf_dir_location,
ufid + COUNTERFACTUAL_FILE_SUFFIX + FILENAME_EXTENSION),
sep=TABULAR_DELIMITER)
true_effect = np.mean(gt[HEADER_Y1] - gt[HEADER_Y0])
true_effects[ufid] = true_effect
# Get the population estimates: | HEADER_EFFECT_SIZE | HEADER_CI_LEFT | HEADER_CI_RIGHT |
estimate = estimates.loc[
ufid, :] # No need to "try:" due to content assertion above
# Calculate the sufficient statistics:
ratio[ufid] = (estimate[HEADER_EFFECT_SIZE] + EPSILON) / (true_effect +
EPSILON)
bias[ufid] = estimate[HEADER_EFFECT_SIZE] - true_effect
ci_size[ufid] = estimate[HEADER_CI_RIGHT] - estimate[
HEADER_CI_LEFT] # right - left -> non-negative
coverage[ufid] = estimate[HEADER_CI_LEFT] <= true_effect <= estimate[
HEADER_CI_RIGHT]
# Save the size of the current dataset:
dataset_sizes[ufid] = gt.index.size
dataset_sizes = dataset_sizes.astype(int) # type: pd.Series
# Calculate metrics
enormse = 1.0 - ratio # type: pd.Series
encis = ci_size / (true_effects.abs() + EPSILON) # type: pd.Series
cic = bias.abs() / ci_size
# Aggregate by sizes:
enormse_by_size = enormse.pow(2).groupby(by=dataset_sizes).mean().pow(0.5)
rmse_by_size = bias.pow(2).groupby(by=dataset_sizes).mean().pow(0.5)
bias_by_size = bias.groupby(by=dataset_sizes).mean()
coverage_by_size = coverage.groupby(by=dataset_sizes).mean()
encis_by_size = encis.groupby(by=dataset_sizes).mean()
cic_by_size = cic.groupby(by=dataset_sizes).mean()
results = pd.Series()
if dataset_sizes.nunique() == 1:
# return the by_sizes, they are enough since there's one size so just extract the scalar value they hold
results["enormse"] = enormse_by_size.iloc[0]
results["rmse"] = rmse_by_size.iloc[0]
results["bias"] = bias_by_size.iloc[0]
results["coverage"] = coverage_by_size.iloc[0]
results["encis"] = encis_by_size.iloc[0]
results["cic"] = cic_by_size.iloc[0]
else:
# weighted_sum = lambda x, w: x.mul(w).sum() / w.sum()
def weighted_sum(x, w):
return x.mul(w).sum() / w.sum()
# Calculate the Weights for aggregation:
weights = __get_weights(dataset_sizes)
# Aggregate
results["enormse"] = np.sqrt(
weighted_sum(enormse_by_size.pow(2), weights))
results["rmse"] = np.sqrt(weighted_sum(rmse_by_size.pow(2), weights))
results["bias"] = weighted_sum(bias_by_size, weights)
results["coverage"] = weighted_sum(coverage_by_size, weights)
results["encis"] = weighted_sum(encis_by_size, weights)
results["cic"] = weighted_sum(cic_by_size, weights)
results = results.append(enormse_by_size.add_prefix("enormse_"))
return results
def opt_nonempty_str_param(obj, param_name, default=None):
if obj is not None and not isinstance(obj, string_types):
raise_with_traceback(_param_type_mismatch_exception(obj, str, param_name))
return default if obj is None or obj == '' else obj
def opt_int_param(obj, param_name, default=None):
if obj is not None and not isinstance(obj, integer_types):
raise_with_traceback(
_param_type_mismatch_exception(obj, int, param_name))
return default if obj is None else obj
def int_param(obj, param_name):
if not isinstance(obj, integer_types):
raise_with_traceback(
_param_type_mismatch_exception(obj, int, param_name))
return obj
def opt_callable_param(obj, param_name, default=None):
if obj is not None and not callable(obj):
raise_with_traceback(_not_callable_exception(obj, param_name))
return default if obj is None else obj
def callable_param(obj, param_name):
if not callable(obj):
raise_with_traceback(_not_callable_exception(obj, param_name))
return obj
def inst_param(obj, param_name, ttype, additional_message=None):
if not isinstance(obj, ttype):
raise_with_traceback(
_param_type_mismatch_exception(
obj, ttype, param_name, additional_message=additional_message))
return obj
def not_implemented(desc):
if not is_str(desc):
raise_with_traceback(CheckError("desc argument must be a string"))
raise_with_traceback(
NotImplementedCheckError("Not implemented: {desc}".format(desc=desc)))
def tuple_param(obj, param_name):
if not isinstance(obj, tuple):
raise_with_traceback(_param_type_mismatch_exception(obj, tuple, param_name))
return obj
def bool_param(obj, param_name):
if not isinstance(obj, bool):
raise_with_traceback(_param_type_mismatch_exception(obj, bool, param_name))
return obj
def opt_int_param(obj, param_name):
if obj is not None and not isinstance(obj, int):
raise_with_traceback(_param_type_mismatch_exception(obj, int, param_name))
return obj
def float_param(obj, param_name):
if not isinstance(obj, float):
raise_with_traceback(
_param_type_mismatch_exception(obj, float, param_name))
return obj
def inst(obj, ttype, desc=None):
if not isinstance(obj, ttype):
raise_with_traceback(_type_mismatch_error(obj, ttype, desc))
return obj
def str_param(obj, param_name):
if not _is_str(obj):
raise_with_traceback(_param_type_mismatch_exception(obj, str, param_name))
return obj
def opt_numeric_param(obj, param_name):
if obj is not None and not isinstance(obj, (int, float)):
raise_with_traceback(
_param_type_mismatch_exception(obj, (int, float), param_name))
return obj
def _score_individual(predictions_location, cf_dir_location):
"""
Scores estimations of treatment effect size on individuals (i.e. the prediction of both outcome under no treatment
and outcome under positive treatment for each individual).
Args:
predictions_location (str): Path to a directory containing tabular files with individual effect estimations
(i.e. prediction of factual and counterfactual outcomes for each individual).
Files must of tabular format
* containing 3 columns: HEADER_IND_IDX, HEADER_Y1, HEADER_Y0.
* delimited by TABULAR_DELIMITER.
* have FILENAME_EXTENSION extension to them.
These global variables specified above can be changed when importing the module.
cf_dir_location (str): Path to a directory containing the counter-factual files (i.e. labeled, ground-truth
data).
Files must be of tabular format
* containing 3 columns: HEADER_IND_IDX, HEADER_Y1, HEADER_Y0.
* delimited by TABULAR_DELIMITER.
* have the suffix specified in COUNTERFACTUAL_FILE_SUFFIX.
* have FILENAME_EXTENSION extension to them.
These global variables specified above can be changed when importing the module.
Returns:
pd.Series: Scores. Where Series' Index is the metric name and the value is the evaluation of that metric.
"""
ufids = os.listdir(predictions_location)
ufids = [
f.rsplit(FILENAME_EXTENSION)[0] for f in ufids
if f.lower().endswith(FILENAME_EXTENSION)
]
enormse = pd.Series(index=ufids, name="individual_enormse")
rmse = pd.Series(index=ufids, name="individual_rmse")
bias = pd.Series(index=ufids, name="individual_bias")
dataset_sizes = pd.Series(index=ufids, name="size")
for ufid in ufids:
# Get the true effect:
gt = pd.read_csv(os.path.join(
cf_dir_location,
ufid + COUNTERFACTUAL_FILE_SUFFIX + FILENAME_EXTENSION),
index_col=HEADER_IND_IDX,
sep=TABULAR_DELIMITER)
true_effect = gt[HEADER_Y1] - gt[HEADER_Y0]
# Get estimated effect: submission format: N rows: patient_ID | Y0 | Y1
try:
estimates = pd.read_csv(os.path.join(predictions_location,
ufid + FILENAME_EXTENSION),
index_col=HEADER_IND_IDX,
sep=TABULAR_DELIMITER)
except IOError as e: # Python 2 compatible for FileNotFoundError
e.args = (
e.args[0] + "\n\t" + "A prediction might be missing.\n"
"Seems that the file ({fn}) was found in the ground-truth directory but no "
"corresponding estimation was found in your "
"predictions.".format(fn=ufid), ) + e.args[1:]
raise_with_traceback(e)
estimated_effect = estimates[HEADER_Y1] - estimates[HEADER_Y0]
# Evaluate:
individual_bias = estimated_effect - true_effect
bias[ufid] = individual_bias.mean()
rmse[ufid] = individual_bias.pow(2).mean()
# enormse[ufid] = np.mean((individual_bias) / (true_effect + EPSILON) ** 2)
enormse[ufid] = np.mean(
(1 - ((estimated_effect + EPSILON) / (true_effect + EPSILON)))**2)
# Save the size of the current dataset:
dataset_sizes[ufid] = gt.index.size
dataset_sizes = dataset_sizes.astype(int) # type: pd.Series
enormse_by_size = enormse.groupby(by=dataset_sizes).mean().pow(0.5)
rmse_by_size = rmse.groupby(by=dataset_sizes).mean().pow(0.5)
bias_by_size = bias.groupby(by=dataset_sizes).mean()
results = pd.Series()
if dataset_sizes.nunique() == 1:
results["enormse"] = enormse_by_size.iloc[0]
results["rmse"] = rmse_by_size.iloc[0]
results["bias"] = bias_by_size.iloc[0]
else:
weights = __get_weights(dataset_sizes)
results["enormse"] = np.sqrt(
enormse_by_size.pow(2).mul(weights).sum() / weights.sum())
results["rmse"] = np.sqrt(
rmse_by_size.pow(2).mul(weights).sum() / weights.sum())
results["bias"] = bias_by_size.mul(weights).sum() / weights.sum()
results = results.append(enormse_by_size.add_prefix("enormse_"))
return results
def failed(desc):
if not _is_str(desc):
raise_with_traceback(CheckError('desc argument must be a string'))
raise_with_traceback(CheckError('Failure condition: {desc}'.format(desc=desc)))
def find_direct_metabolites(model, reaction, tolerance=1E-06):
"""
Return list of possible direct biomass precursor metabolites.
The term direct metabolites describes metabolites that are involved only
in either transport and/or boundary reactions, AND the biomass reaction(s),
but not in any purely metabolic reactions.
Parameters
----------
model : cobra.Model
The metabolic model under investigation.
reaction : cobra.Reaction
The biomass reaction of the model under investigation.
tolerance : float, optional
Tolerance below which values will be regarded as zero.
Returns
-------
list
Metabolites that qualify as direct metabolites i.e. biomass precursors
that are taken up to be consumed by the biomass reaction only.
"""
biomass_rxns = set(helpers.find_biomass_reaction(model))
tra_bou_bio_rxns = helpers.find_interchange_biomass_reactions(
model, biomass_rxns)
try:
precursors = find_biomass_precursors(model, reaction)
main_comp = helpers.find_compartment_id_in_model(model, 'c')
ext_space = helpers.find_compartment_id_in_model(model, 'e')
except KeyError:
LOGGER.error("Failed to properly identify cytosolic and extracellular "
"compartments.")
raise_with_traceback(
KeyError("The cytosolic and/or extracellular "
"compartments could not be identified."))
except RuntimeError:
LOGGER.error("Failed to properly identify cytosolic and extracellular "
"compartments.")
raise_with_traceback(
RuntimeError("The cytosolic and/or extracellular "
"compartments could not be "
"identified."))
else:
tra_bou_bio_mets = [
met for met in precursors
if met.reactions.issubset(tra_bou_bio_rxns)
]
rxns_of_interest = set([
rxn for met in tra_bou_bio_mets for rxn in met.reactions
if rxn not in biomass_rxns
])
solution = model.optimize(raise_error=True)
if np.isclose(solution.objective_value, 0, atol=tolerance):
LOGGER.error("Failed to generate a non-zero objective value with "
"flux balance analysis.")
raise OptimizationError(
"The flux balance analysis on this model returned an "
"objective value of zero. Make sure the model can "
"grow! Check if the constraints are not too strict!")
tra_bou_bio_fluxes = {r: solution[r.id] for r in rxns_of_interest}
met_flux_sum = {m: 0 for m in tra_bou_bio_mets}
return detect_false_positive_direct_metabolites(tra_bou_bio_mets,
biomass_rxns, main_comp,
ext_space,
tra_bou_bio_fluxes,
met_flux_sum)
def subclass_param(obj, param_name, superclass):
type_param(obj, param_name)
if not issubclass(obj, superclass):
raise_with_traceback(_param_subclass_mismatch_exception(obj, superclass, param_name))
return obj
def str_param(obj: Any, param_name: str) -> str:
if not isinstance(obj, str):
raise_with_traceback(
_param_type_mismatch_exception(obj, str, param_name))
return obj
def opt_type_param(obj, param_name, default=None):
if obj is not None and not isinstance(obj, type):
raise_with_traceback(_not_type_param_subclass_mismatch_exception(obj, param_name))
return obj if obj is not None else default
def int_param(obj: Any, param_name: str) -> int:
if not isinstance(obj, int):
raise_with_traceback(
_param_type_mismatch_exception(obj, int, param_name))
return obj
def type_param(obj, param_name):
if not isinstance(obj, type):
raise_with_traceback(_not_type_param_subclass_mismatch_exception(obj, param_name))
return obj
def param_invariant(condition, param_name, desc=None):
if not condition:
raise_with_traceback(_param_invariant_exception(param_name, desc))
def exec_one_test(self, test_name, test_method):
"""Executes one test and update test results.
Executes setup_test, the test method, and teardown_test; then creates a
records.TestResultRecord object with the execution information and adds
the record to the test class's test results.
Args:
test_name: string, Name of the test.
test_method: function, The test method to execute.
"""
tr_record = records.TestResultRecord(test_name, self.TAG)
tr_record.test_begin()
self.current_test_info = runtime_test_info.RuntimeTestInfo(
test_name, self.log_path, tr_record)
expects.recorder.reset_internal_states(tr_record)
logging.info('%s %s', TEST_CASE_TOKEN, test_name)
# Did teardown_test throw an error.
teardown_test_failed = False
try:
try:
try:
self._setup_test(test_name)
except signals.TestFailure as e:
raise_with_traceback(signals.TestError(
e.details, e.extras))
test_method()
except signals.TestPass:
raise
except Exception:
logging.exception('Exception occurred in %s.',
self.current_test_name)
raise
finally:
before_count = expects.recorder.error_count
try:
self._teardown_test(test_name)
except signals.TestAbortSignal:
raise
except Exception as e:
logging.exception(e)
tr_record.test_error()
tr_record.add_error(STAGE_NAME_TEARDOWN_TEST, e)
teardown_test_failed = True
else:
# Check if anything failed by `expects`.
if before_count < expects.recorder.error_count:
teardown_test_failed = True
except (signals.TestFailure, AssertionError) as e:
tr_record.test_fail(e)
except signals.TestSkip as e:
# Test skipped.
tr_record.test_skip(e)
except signals.TestAbortSignal as e:
# Abort signals, pass along.
tr_record.test_fail(e)
raise e
except signals.TestPass as e:
# Explicit test pass.
tr_record.test_pass(e)
except Exception as e:
# Exception happened during test.
tr_record.test_error(e)
else:
# No exception is thrown from test and teardown, if `expects` has
# error, the test should fail with the first error in `expects`.
if expects.recorder.has_error and not teardown_test_failed:
tr_record.test_fail()
# Otherwise the test passed.
elif not teardown_test_failed:
tr_record.test_pass()
finally:
tr_record.update_record()
try:
if tr_record.result in (
records.TestResultEnums.TEST_RESULT_ERROR,
records.TestResultEnums.TEST_RESULT_FAIL):
self._exec_procedure_func(self._on_fail, tr_record)
elif tr_record.result == records.TestResultEnums.TEST_RESULT_PASS:
self._exec_procedure_func(self._on_pass, tr_record)
elif tr_record.result == records.TestResultEnums.TEST_RESULT_SKIP:
self._exec_procedure_func(self._on_skip, tr_record)
finally:
logging.info(RESULT_LINE_TEMPLATE, tr_record.test_name,
tr_record.result)
self.results.add_record(tr_record)
self.summary_writer.dump(tr_record.to_dict(),
records.TestSummaryEntryType.RECORD)
self.current_test_info = None
self.current_test_name = None
