Black and flake8 current code (#4)

* remove old_petab files for simple model

.flake8

@@ -0,0 +1,3 @@
+[flake8]
+per-file-ignores =
+    */__init__.py:F401,F403

fiddy/__init__.py

@@ -1,5 +1,3 @@
-# noqa: F401, F403
-
 from .constants import *
 from . import difference
 from . import function

fiddy/constants.py

@@ -1,11 +1,15 @@
 from enum import Enum
-from typing import Callable, Iterable, Union
+from typing import Callable, Union
 
 import numpy as np
 
 __all__ = [
-    'TYPE_DIMENSION', 'TYPE_OUTPUT', 'TYPE_POINT', 'TYPE_FUNCTION',
-    'Difference', 'GradientCheckMethod'
+    "TYPE_DIMENSION",
+    "TYPE_FUNCTION",
+    "TYPE_OUTPUT",
+    "TYPE_POINT",
+    "Difference",
+    "GradientCheckMethod",
 ]
 
 # Currently only 1D arrays are supported.
@@ -15,18 +19,18 @@
 #      - `.gradient_check` classes
 #      ... or just flatten
 TYPE_DIMENSION = int
-TYPE_POINT = np.ndarray
 TYPE_OUTPUT = Union[float, int]
+TYPE_POINT = np.ndarray
 TYPE_FUNCTION = Callable[[TYPE_POINT], TYPE_OUTPUT]
 
 
 class Difference(str, Enum):
-    FORWARD = 'forward'
-    BACKWARD = 'backward'
-    CENTRAL = 'central'
+    BACKWARD = "backward"
+    CENTRAL = "central"
+    FORWARD = "forward"
 
 
 class GradientCheckMethod(str, Enum):
-    FORWARD = Difference.FORWARD
     BACKWARD = Difference.BACKWARD
     CENTRAL = Difference.CENTRAL
+    FORWARD = Difference.FORWARD

fiddy/difference.py

@@ -58,4 +58,4 @@ def central(
     Returns:
         The central difference.
     """
-    return function(point + step/2) - function(point - step/2)
+    return function(point + step / 2) - function(point - step / 2)

fiddy/extensions/__init__.py

@@ -0,0 +1,3 @@
+"""Extensions, e.g. for compatibility with other tools."""
+
+from . import amici

fiddy/extensions/amici/amici.py

@@ -1,7 +1,6 @@
 from functools import partial
 from typing import Any, Callable, Dict, List, Tuple
 
-import amici
 import numpy as np
 from amici.petab_objective import LLH, SLLH
 import petab
@@ -27,9 +26,9 @@ def transform_gradient_log10(gradient_value, parameter_value):
 
 
 transforms = {
-    'lin': transform_gradient_lin,
-    'log': transform_gradient_log,
-    'log10': transform_gradient_log10,
+    "lin": transform_gradient_lin,
+    "log": transform_gradient_log,
+    "log10": transform_gradient_log10,
 }
 
 
@@ -72,8 +71,9 @@ def simulate_petab_to_cached_functions(
         for parameter_id in parameter_ids
     ]
 
-    simulate_petab_partial = \
-        partial(simulate_petab, petab_problem=petab_problem, *args, **kwargs)
+    simulate_petab_partial = partial(
+        simulate_petab, petab_problem=petab_problem, *args, **kwargs
+    )
 
     def simulate_petab_full(point: TYPE_POINT):
         problem_parameters = dict(zip(parameter_ids, point))
@@ -90,14 +90,15 @@ def function(point: TYPE_POINT):
 
     def gradient(point: TYPE_POINT) -> TYPE_POINT:
         result = simulate_petab_full_cached(point)
-        sllh = np.array([
-            gradient_transformations[parameter_index](
-                gradient_value=result[SLLH][parameter_id],
-                parameter_value=point[parameter_index],
-            )
-            for parameter_index, parameter_id in enumerate(parameter_ids)
-        ])
+        sllh = np.array(
+            [
+                gradient_transformations[parameter_index](
+                    gradient_value=result[SLLH][parameter_id],
+                    parameter_value=point[parameter_index],
+                )
+                for parameter_index, parameter_id in enumerate(parameter_ids)
+            ]
+        )
         return sllh
 
     return function, gradient
-

fiddy/function.py

@@ -7,14 +7,15 @@
 
 
 default_memory_kwargs = {
-    'location': 'cache_fiddy',
-    'verbose': 0,
+    "location": "cache_fiddy",
+    "verbose": 0,
 }
 ram_cache_parent_path = Path("/dev/shm")
 
 
 class Function:
     """Wrapper for functions."""
+
     def __init__(
         self,
         function: TYPE_FUNCTION,
@@ -60,10 +61,11 @@ def __init__(
             ram_cache:
                 Whether to cache in RAM. If `False`, disk is used instead.
         """
-        self.cache_path = \
-            kwargs.get('location', default_memory_kwargs['location'])
+        self.cache_path = kwargs.get(
+            "location", default_memory_kwargs["location"]
+        )
         if ram_cache:
-            if 'location' in kwargs:
+            if "location" in kwargs:
                 raise ValueError(
                     "Do not supply a location when using `ram_cache`."
                 )
@@ -72,10 +74,11 @@ def __init__(
                     "The standard Linux shared memory location '/dev/shm' "
                     "does not exist."
                 )
-            self.cache_path = \
-                ram_cache_parent_path / default_memory_kwargs['location']
+            self.cache_path = (
+                ram_cache_parent_path / default_memory_kwargs["location"]
+            )
         self.cache_path = Path(self.cache_path).resolve()
-        kwargs['location'] = str(self.cache_path)
+        kwargs["location"] = str(self.cache_path)
 
         memory = joblib.Memory(**{**default_memory_kwargs, **kwargs})
         self.function = memory.cache(function)

fiddy/gradient_check.py

@@ -1,4 +1,3 @@
-import abc
 from dataclasses import dataclass
 from functools import partial
 from typing import Callable, Iterable, List, Tuple
@@ -10,11 +9,8 @@
     TYPE_DIMENSION,
     TYPE_FUNCTION,
     TYPE_POINT,
-    Difference,
     GradientCheckMethod,
 )
-from .misc import numpy_array_to_tuple
-#from .quotient import forward, backward, central
 from .step import dstep
 
 
@@ -27,8 +23,8 @@ def gradient_check(
     stop_at_success: bool = True,
     # TODO or custom Callable
     fd_gradient_method: GradientCheckMethod = None,
-    #atol: float = 1e-2,
-    #rtol: float = 1e-2,
+    # atol: float = 1e-2,  # TODO
+    # rtol: float = 1e-2,  # TODO
     check_protocol: List[Callable[[pd.DataFrame], None]] = None,
     postprocessor_protocol: List[Callable[[pd.DataFrame], None]] = None,
 ) -> Tuple[bool, pd.DataFrame]:
@@ -60,8 +56,8 @@ def gradient_check(
         check_protocol:
             These methods are applied to the results, to perform the checks.
             Defaults to `default_check_protocol`.
-            Methods in this protocol should set the `"success"` column to `True`
-            if the check passes, and put the reason for success in the
+            Methods in this protocol should set the `"success"` column to
+            `True` if the check passes, and put the reason for success in the
             `"success_reason"` column.
         postprocessor_protocol:
             Similar to `check_protocol`, but applied after `check_protocol`.
@@ -77,7 +73,7 @@ def gradient_check(
     if sizes is None:
         sizes = [1e-1, 1e-3, 1e-5, 1e-7, 1e-9]
     if fd_gradient_method is None:
-        fd_gradient_method = 'central'
+        fd_gradient_method = "central"
     if check_protocol is None:
         check_protocol = default_check_protocol
     if postprocessor_protocol is None:
@@ -90,25 +86,27 @@ def gradient_check(
     # Create a method to approximate the gradient. Should only require a
     # step as its only argument (use as kwarg).
     # `fd_gradient_callable` should only require a step to run.
-    if fd_gradient_method == 'forward':
-        fd_gradient_callable = \
-            partial(quotient.forward, function=function, point=point)
-    elif fd_gradient_method == 'backward':
-        fd_gradient_callable = \
-            partial(quotient.backward, function=function, point=point)
-    elif fd_gradient_method == 'central':
-        fd_gradient_callable = \
-            partial(quotient.central, function=function, point=point)
+    if fd_gradient_method == "forward":
+        fd_gradient_callable = partial(
+            quotient.forward, function=function, point=point
+        )
+    elif fd_gradient_method == "backward":
+        fd_gradient_callable = partial(
+            quotient.backward, function=function, point=point
+        )
+    elif fd_gradient_method == "central":
+        fd_gradient_callable = partial(
+            quotient.central, function=function, point=point
+        )
     else:
-        raise NotImplementedError(f'Method: {fd_gradient_method}')
+        raise NotImplementedError(f"Method: {fd_gradient_method}")
 
     for size in sizes:
         for dimension in dimensions:
             step = dstep(point=point, dimension=dimension, size=size)
             test_gradient = fd_gradient_callable(step=step)
             results.append(
                 Result(
-                    #point=numpy_array_to_tuple(point),
                     dimension=dimension,
                     size=size,
                     test_gradient=test_gradient,
@@ -118,8 +116,8 @@ def gradient_check(
             )
 
     results_df = pd.DataFrame(results)
-    results_df['success'] = False
-    results_df['success_reason'] = None
+    results_df["success"] = False
+    results_df["success_reason"] = None
     for check in check_protocol:
         check(results_df)
     if postprocessor_protocol is not None:
@@ -134,7 +132,8 @@ def gradient_check(
     return success, results_df
 
 
-# FIXME refactor to some `gradient.py` where these FD methods can be used to compute gradients.
+# FIXME refactor to some `gradient.py` where these FD methods can be used to
+#       compute gradients.
 #       would result in or require a similar method
 def simplify_results_df(results_df: pd.DataFrame) -> pd.DataFrame:
     """Keep only one row per successful dimension, in the dataframe.
@@ -150,7 +149,7 @@ def simplify_results_df(results_df: pd.DataFrame) -> pd.DataFrame:
             The simplified results.
     """
     dimension_result_dfs = []
-    for dimension, df in results_df.groupby('dimension'):
+    for dimension, df in results_df.groupby("dimension"):
         # If any checks were successful for this dimension, only include the
         # first successful check.
         if df["success"].any():
@@ -167,8 +166,7 @@ def simplify_results_df(results_df: pd.DataFrame) -> pd.DataFrame:
 @dataclass
 class Result:
     """Information about a single finite difference gradient computation."""
-    # """The point at which the gradient was computed."""
-    # point: TYPE_POINT
+
     size: float
     """The size of the step taken."""
     dimension: TYPE_DIMENSION
@@ -183,54 +181,54 @@ class Result:
 
 # FIXME string literals
 def add_absolute_error(results_df):
-    results_df['|aerr|'] = abs(
-        results_df['test_gradient'] - results_df['expected_gradient']
+    results_df["|aerr|"] = abs(
+        results_df["test_gradient"] - results_df["expected_gradient"]
     )
 
 
 def check_absolute_error(results_df, tolerance: float = 1e-2):
-    success = results_df['|aerr|'] < tolerance
-    set_success(results_df, success, reason='|aerr|')
+    success = results_df["|aerr|"] < tolerance
+    set_success(results_df, success, reason="|aerr|")
 
 
 def add_relative_error(results_df):
-    if '|aerr|' not in results_df.columns:
+    if "|aerr|" not in results_df.columns:
         add_absolute_error(results_df)
-    epsilon = results_df['|aerr|'].min() * 1e-10
+    epsilon = results_df["|aerr|"].min() * 1e-10
 
-    results_df['|rerr|'] = abs(
-        results_df['|aerr|'] / (results_df['expected_gradient'] + epsilon)
+    results_df["|rerr|"] = abs(
+        results_df["|aerr|"] / (results_df["expected_gradient"] + epsilon)
     )
 
 
 def check_relative_error(results_df, tolerance: float = 1e-2):
-    success = results_df['|rerr|'] < tolerance
-    set_success(results_df, success, reason='|rerr|')
+    success = results_df["|rerr|"] < tolerance
+    set_success(results_df, success, reason="|rerr|")
 
 
 def set_success(results_df: pd.DataFrame, success: pd.Series, reason: str):
-    new_success = success & ~results_df['success']
-    results_df['success'] = results_df['success'] | new_success
-    results_df['success_reason'].mask(
+    new_success = success & ~results_df["success"]
+    results_df["success"] = results_df["success"] | new_success
+    results_df["success_reason"].mask(
         new_success,
         reason,
         inplace=True,
     )
 
 
 default_check_protocol = [
-    #set_all_failed,
+    # set_all_failed,
     add_absolute_error,
     add_relative_error,
     check_absolute_error,
     check_relative_error,
 ]
 
 
-def keep_lowest_error(results_df, error='|rerr|'):
+def keep_lowest_error(results_df, error="|rerr|"):
     keep_indices = []
 
-    for dimension, df in results_df.groupby('dimension'):
+    for dimension, df in results_df.groupby("dimension"):
         # Keep best success from each dimension.
         if df["success"].any():
             keep_index = df.loc[df["success"]][error].idxmin()

fiddy/misc.py

@@ -18,9 +18,6 @@ def numpy_array_to_tuple(array: np.ndarray) -> Tuple:
     # Iterate over the array, turning iterable
     # elements also into tuples, recursively.
     try:
-        return tuple(
-            numpy_array_to_tuple(element)
-            for element in array
-        )
+        return tuple(numpy_array_to_tuple(element) for element in array)
     except TypeError:
         return array

fiddy/quotient.py

@@ -1,5 +1,3 @@
-from typing import Optional
-
 import numpy as np
 
 from .constants import (