Update evaluator

src/scgenerator/const.py

@@ -19,6 +19,8 @@ def pbar_format(worker_id: int) -> dict[str, Any]:
         )
 
 
+INF = float("inf")
+
 SPEC1_FN = "spectrum_{}.npy"
 SPECN_FN1 = "spectra_{}.npy"
 SPEC1_FN_N = "spectrum_{}_{}.npy"

src/scgenerator/errors.py

@@ -30,15 +30,3 @@ class MissingParameterError(Exception):
 
 class IncompleteDataFolderError(FileNotFoundError):
     pass
-
-
-class EvaluatorError(Exception):
-    pass
-
-
-class NoDefaultError(EvaluatorError):
-    pass
-
-
-class OperatorError(EvaluatorError):
-    pass

src/scgenerator/evaluator.py

@@ -1,25 +1,96 @@
+from __future__ import annotations
+
 import itertools
-from collections import defaultdict
-from dataclasses import dataclass
-from typing import Any, Callable, Optional, Union
+import traceback
+from collections import ChainMap, defaultdict
+from typing import Any, Callable, MutableMapping, NamedTuple, Optional, Type, Union
 
 import numpy as np
 
 from scgenerator import math, operators, utils
-from scgenerator.const import MANDATORY_PARAMETERS
-from scgenerator.errors import EvaluatorError, NoDefaultError
+from scgenerator.const import INF, MANDATORY_PARAMETERS
 from scgenerator.physics import fiber, materials, plasma, pulse, units
 from scgenerator.utils import _mock_function, func_rewrite, get_arg_names, get_logger
 
 
+class ErrorRecord(NamedTuple):
+    error: Exception
+    traceback: str
+
+
+class EvaluatorError(Exception):
+    pass
+
+
+class NoValidRuleError(EvaluatorError):
+    def __init__(self, target: str):
+        ...
+
+
+class CyclicDependencyError(EvaluatorError):
+    def __init__(self, target: str, lookup_stack: list[str]):
+        ...
+
+
+class AllRulesExhaustedError(EvaluatorError):
+    def __init__(self, target: str):
+        ...
+
+
+class EvaluatorErrorTree:
+    target: str
+    all_errors: dict[Type, ErrorRecord]
+
+    level: int
+
+    def __init__(self, target: str):
+        self.target = target
+        self.all_errors = {}
+        self.level = 0
+
+    def __repr__(self) -> str:
+        return f"{self.__class__.__name__}(target={self.target!r})"
+
+    def push(self):
+        self.level += 1
+
+    def pop(self):
+        self.level -= 1
+
+    def append(self, error: Exception):
+        tr = traceback.format_exc().splitlines()
+        tr = "\n".join(tr[:1] + tr[3:])  # get rid of the `append` frame
+        self.all_errors[type(error)] = ErrorRecord(error, tr)
+
+    def compile_error(self) -> EvaluatorError:
+        raise EvaluatorError(f"Couldn't compute {self.target}.")
+
+    def summary(self) -> dict[str, int]:
+        return {k.__name__: len(v) for k, v in self.all_errors.items()}
+
+    def format_summary(self) -> str:
+        types = [f"{v} errors of type {k!r}" for k, v in self.summary().items()]
+        return ", ".join(types[:-2] + [" and ".join(types[-2:])])
+
+    def details(self) -> str:
+        ...
+
+
 class Rule:
+    targets: dict[str, int]
+    func: Callable
+    args: list[str]
+    conditions: dict[str, Any]
+
+    mock_func: Callable
+
     def __init__(
         self,
         target: Union[str, list[Optional[str]]],
         func: Callable,
-        args: list[str] = None,
-        priorities: Union[int, list[int]] = None,
-        conditions: dict[str, str] = None,
+        args: list[str] | None = None,
+        priorities: Union[int, list[int]] | None = None,
+        conditions: dict[str, str] | None = None,
     ):
         targets = list(target) if isinstance(target, (list, tuple)) else [target]
         self.func = func
@@ -43,6 +114,16 @@ class Rule:
             f"{self.func.__name__} --> [{', '.join(self.targets)}]"
         )
 
+    def __eq__(self, other: Rule) -> bool:
+        return self.func == other.func
+
+    def pretty_format(self) -> str:
+        func_name_elements = self.func_name.split(".")
+        targets = list(self.targets)
+
+        arg_size = max(self.args, key=len)
+        func_size = max(func_name_elements, key=len)
+
     @property
     def func_name(self) -> str:
         return f"{self.func.__module__}.{self.func.__name__}"
@@ -58,7 +139,7 @@ class Rule:
         priorities: Union[int, list[int]] = None,
     ) -> list["Rule"]:
         """
-        given a function that doesn't need all its keyword arguemtn specified, will
+        given a function that doesn't need all its keyword arguments specified, will
         return a list of Rule obj, one for each combination of n_var specified kwargs
 
         Parameters
@@ -97,15 +178,21 @@ class Rule:
         return rules
 
 
-@dataclass
-class EvalStat:
-    priority: float = np.inf
+class EvaluatedValue(NamedTuple):
+    value: Any
+    priority: float = INF
     rule: Rule = None
 
 
 class Evaluator:
     defaults: dict[str, Any] = {}
 
+    rules: dict[str, list[Rule]]
+    main_map: dict[str, EvaluatedValue]
+    param_chain_map: MutableMapping[str, EvaluatedValue]
+    lookup_stack: list[str]
+    failed_rules: dict[str, list[Rule]]
+
     @classmethod
     def default(cls, full_field: bool = False) -> "Evaluator":
         evaluator = cls()
@@ -124,28 +211,33 @@ class Evaluator:
         evaluator.set(**params)
         for target in MANDATORY_PARAMETERS:
             evaluator.compute(target, check_only=check_only)
-        return evaluator.params
+        return evaluator.main_map
 
     @classmethod
     def register_default_param(cls, key, value):
         cls.defaults[key] = value
 
-    def __init__(self):
-        self.rules: dict[str, list[Rule]] = defaultdict(list)
-        self.params = {}
-        self.__curent_lookup: list[str] = []
-        self.__failed_rules: dict[str, list[Rule]] = defaultdict(list)
-        self.eval_stats: dict[str, EvalStat] = defaultdict(EvalStat)
+    def __init__(self, *rules: Rule):
+        self.rules = defaultdict(list)
+        self.main_map = {}
         self.logger = get_logger(__name__)
 
+        self.failed_rules = defaultdict(list)
+
+        self.append(*rules)
+
+    def __getitem__(self, key: str) -> Any:
+        return self.main_map[key].value
+
     def append(self, *rule: Rule):
         for r in rule:
             for t in r.targets:
-                if t is not None:
-                    self.rules[t].append(r)
-                    self.rules[t].sort(key=lambda el: el.targets[t], reverse=True)
+                if t is None:
+                    continue
+                self.rules[t].append(r)
+                self.rules[t].sort(key=lambda el: el.targets[t], reverse=True)
 
-    def set(self, dico: dict[str, Any] = None, **params: Any):
+    def set(self, **params: Any):
         """
         sets the internal set of parameters
 
@@ -157,18 +249,11 @@ class Evaluator:
         params : Any
             if dico is None, update the internal dict of parameters with params
         """
-        if dico is None:
-            dico = params
-            self.params.update(dico)
-        else:
-            self.reset()
-            self.params = dico
-        for k in dico:
-            self.eval_stats[k].priority = np.inf
+        for k, v in params.items():
+            self.main_map[k] = EvaluatedValue(v, np.inf)
 
     def reset(self):
-        self.params = {}
-        self.eval_stats = defaultdict(EvalStat)
+        self.main_map = {}
 
     def compute(self, target: str, check_only=False) -> Any:
         """
@@ -191,99 +276,115 @@ class Evaluator:
         KeyError
             there is no saved rule for the target
         """
-        value = self.params.get(target)
-        if value is None:
-            prefix = "\t" * len(self.__curent_lookup)
-            # Avoid cycles
-            if target in self.__curent_lookup:
-                raise EvaluatorError(
-                    "cyclic dependency detected : "
-                    f"{target!r} seems to depend on itself, please provide "
-                    f"a value for at least one variable in {self.__curent_lookup!r}. "
-                    + self.attempted_rules_str(target)
-                )
-            else:
-                self.__curent_lookup.append(target)
-
-            if len(self.rules[target]) == 0:
-                error = EvaluatorError(
-                    f"no rule for {target}, trying to evaluate {self.__curent_lookup!r}"
-                )
-            else:
-                error = None
-
-            # try every rule until one succeeds
-            for ii, rule in enumerate(filter(self.validate_condition, self.rules[target])):
-                self.logger.debug(
-                    prefix + f"attempt {ii+1} to compute {target}, this time using {rule!r}"
-                )
-                try:
-                    args = [self.compute(k, check_only=check_only) for k in rule.args]
-                    if check_only:
-                        returned_values = rule.mock_func(*args)
-                    else:
-                        returned_values = rule.func(*args)
-                    if len(rule.targets) == 1:
-                        returned_values = [returned_values]
-                    for (param_name, param_priority), returned_value in zip(
-                        rule.targets.items(), returned_values
-                    ):
-                        if (
-                            param_name == target
-                            or param_name not in self.params
-                            or self.eval_stats[param_name].priority < param_priority
-                        ):
-                            if check_only:
-                                success_str = f"able to compute {param_name} "
-                            else:
-                                v_str = format(returned_value).replace("\n", "")
-                                success_str = f"computed {param_name}={v_str} "
-                            self.logger.debug(
-                                prefix
-                                + success_str
-                                + f"using {rule.func.__name__} from {rule.func.__module__}"
-                            )
-                            self.set_value(param_name, returned_value, param_priority, rule)
-                        if param_name == target:
-                            value = returned_value
-                    break
-                except EvaluatorError as e:
-                    error = e
-                    self.logger.debug(
-                        prefix + f"error using {rule.func.__name__} : {str(error).strip()}"
-                    )
-                    self.__failed_rules[target].append(rule)
-                    continue
-                except Exception as e:
-                    raise type(e)(f"error while evaluating {target!r}")
-            else:
-                default = self.defaults.get(target)
-                if default is None:
-                    error = error or NoDefaultError(
-                        prefix
-                        + f"No default provided for {target}. Current lookup cycle : {self.__curent_lookup!r}. "
-                        + self.attempted_rules_str(target)
-                    )
-                else:
-                    value = default
-                    self.logger.info(prefix + f"using default value of {value} for {target}")
-                    self.set_value(target, value, 0, None)
-            last_target = self.__curent_lookup.pop()
-            assert target == last_target
-            self.__failed_rules[target] = []
-
-            if value is None and error is not None:
-                raise error
-
+        errors = EvaluatorErrorTree(target)
+        param_chain_map = ChainMap(self.main_map)
+        try:
+            value = self._compute(target, check_only, param_chain_map, [], errors)
+        except EvaluatorError as e:
+            errors.append(e)
+            raise errors.compile_error() from None
+        self.merge_chain_map(param_chain_map)
         return value
 
-    def __getitem__(self, key: str) -> Any:
-        return self.params[key]
+    def _compute(
+        self,
+        target: str,
+        check_only: bool,
+        param_chain_map: MutableMapping[str, EvaluatedValue],
+        lookup_stack: list[str],
+        errors: EvaluatorErrorTree,
+    ) -> Any:
+        errors.push()
+        if target in param_chain_map:
+            return param_chain_map[target].value
+        if target not in self.rules or len(self.rules[target]) == 0:
+            raise NoValidRuleError(target)
+        if target in lookup_stack:
+            raise CyclicDependencyError(target, lookup_stack.copy())
 
-    def set_value(self, key: str, value: Any, priority: int, rule: Rule):
-        self.params[key] = value
-        self.eval_stats[key].priority = priority
-        self.eval_stats[key].rule = rule
+        lookup_stack.append(target)
+        base_cm_length = len(param_chain_map.maps)
+        for rule in self.rules[target]:
+            values = self.apply_rule(rule, check_only, param_chain_map, lookup_stack, errors)
+            if self.valid_values(values, rule, check_only, param_chain_map, lookup_stack, errors):
+                break
+        lookup_stack.pop()
+
+        if values is None:
+            if target in self.defaults:
+                values = {target: self.defaults[target]}
+            else:
+                self.clear_chain_map(param_chain_map, base_cm_length)
+                raise AllRulesExhaustedError(target)
+
+        param_chain_map.maps.insert(0, values)
+        errors.pop()
+        return values[target].value
+
+    def valid_values(
+        self,
+        values: dict[str, EvaluatedValue] | None,
+        rule: Rule,
+        check_only: bool,
+        param_chain_map: MutableMapping[str, EvaluatedValue],
+        lookup_stack: list[str],
+        errors: EvaluatorErrorTree,
+    ) -> bool:
+        if values is None:
+            return False
+        if check_only:
+            return True
+
+        for arg, target_value in rule.conditions.items():
+            value = self._compute(arg, False, param_chain_map, lookup_stack, errors)
+            if value != target_value:
+                return False
+
+        return True
+
+    def apply_rule(
+        self,
+        rule: Rule,
+        check_only: bool,
+        param_chain_map: MutableMapping[str, EvaluatedValue],
+        lookup_stack: list[str],
+        errors: EvaluatorErrorTree,
+    ) -> dict[str, EvaluatedValue] | None:
+        try:
+            for arg in rule.args:
+                self._compute(arg, check_only, param_chain_map, lookup_stack, errors)
+        except Exception as e:
+            errors.append(e)
+            return None
+
+        args = [param_chain_map[k].value for k in rule.args]
+
+        func = rule.mock_func if check_only else rule.func
+
+        try:
+            values = func(*args)
+        except Exception as e:
+            self.record_error(e)
+            return None
+
+        if not isinstance(values, tuple):
+            values = (values,)
+
+        return {k: EvaluatedValue(v, rule.targets[k], rule) for k, v in zip(rule.targets, values)}
+
+    def merge_chain_map(self, param_chain_map: dict[str, EvaluatedValue]):
+        while len(param_chain_map.maps) > 1:
+            params = param_chain_map.maps.pop(0)
+            for k, v in params.items():
+                target_priority = self.main_map[k].priority if k in self.main_map else -INF
+                if v.priority > target_priority:
+                    self.main_map[k] = v
+
+    def clear_chain_map(
+        self, param_chain_map: MutableMapping[str, EvaluatedValue], target_size: int
+    ):
+        while len(param_chain_map.maps) > target_size:
+            param_chain_map.pop(0)
 
     def validate_condition(self, rule: Rule) -> bool:
         try:
@@ -292,7 +393,7 @@ class Evaluator:
             return False
 
     def attempted_rules_str(self, target: str) -> str:
-        rules = ", ".join(str(r) for r in self.__failed_rules[target])
+        rules = ", ".join(str(r) for r in self.failed_rules[target])
         if len(rules) == 0:
             return ""
         return "attempted rules : " + rules

src/scgenerator/parameter.py

@@ -14,8 +14,7 @@ import numpy as np
 
 from scgenerator import utils
 from scgenerator.const import MANDATORY_PARAMETERS, __version__
-from scgenerator.errors import EvaluatorError
-from scgenerator.evaluator import Evaluator
+from scgenerator.evaluator import Evaluator, EvaluatorError
 from scgenerator.io import DatetimeEncoder, decode_datetime_hook
 from scgenerator.operators import Qualifier, SpecOperator
 from scgenerator.utils import update_path_name
@@ -475,7 +474,7 @@ class Parameters:
 
     def get_evaluator(self):
         evaluator = Evaluator.default(self.full_field)
-        evaluator.set(self._param_dico.copy())
+        evaluator.set(**self._param_dico)
         return evaluator
 
     def dump_dict(self, add_metadata=True) -> dict[str, Any]:
@@ -546,14 +545,15 @@ class Parameters:
             ) from None
         if exhaustive:
             for p in self._p_names:
-                if p not in evaluator.params:
+                if p not in evaluator.main_map:
                     try:
                         evaluator.compute(p)
                     except Exception:
                         pass
         computed = self.__class__(
-            **{k: v for k, v in evaluator.params.items() if k in self._p_names}
+            **{k: v.value for k, v in evaluator.main_map.items() if k in self._p_names}
         )
+        computed._frozen = True
         return computed
 
     def pretty_str(self, params: Iterable[str] = None, exclude=None) -> str:

tests/test_evaluator.py

@@ -0,0 +1,23 @@
+import numpy as np
+import pytest
+
+from scgenerator.evaluator import Evaluator, Rule
+
+
+@pytest.fixture
+def disk_rules() -> list[Rule]:
+    return [
+        Rule("radius", lambda diameter: diameter / 2),
+        Rule("diameter", lambda radius: radius * 2),
+        Rule("diameter", lambda perimeter: perimeter / np.pi),
+        Rule("perimeter", lambda diameter: diameter * np.pi),
+        Rule("area", lambda radius: np.pi * radius**2),
+        Rule("radius", lambda area: np.sqrt(area / np.pi)),
+    ]
+
+
+def test_simple(disk_rules: list[Rule]):
+    evaluator = Evaluator(*disk_rules, Rule("radius", lambda lol: lol * 3))
+    evaluator.set()
+
+    assert evaluator.compute("area") == pytest.approx(78.53981633974483)