Fixes for long-term runs (#924)

* The unlimited number of generations is configured for composer (to satisfy any timeout)
* The same for tuner's iterations
* Early stopping reason logging

fedot/api/api_utils/api_composer.py

@@ -29,7 +29,6 @@
 from fedot.core.pipelines.tuning.tuner_builder import TunerBuilder
 from fedot.core.pipelines.tuning.unified import PipelineTuner
 from fedot.core.pipelines.verification import rules_by_task
-from fedot.core.repository.operation_types_repository import get_operations_for_task
 from fedot.core.repository.pipeline_operation_repository import PipelineOperationRepository
 from fedot.core.repository.quality_metrics_repository import MetricsRepository, MetricType, MetricsEnum
 from fedot.core.repository.tasks import Task, TaskTypesEnum
@@ -197,8 +196,9 @@ def compose_fedot_model(self, api_params: dict, composer_params: dict, tuning_pa
                 assumption_handler.fit_assumption_and_check_correctness(initial_assumption[0],
                                                                         pipelines_cache=self.pipelines_cache,
                                                                         preprocessing_cache=self.preprocessing_cache)
+
         self.log.message(
-            f'Initial pipeline was fitted in {round(self.timer.assumption_fit_spend_time.total_seconds())} sec.')
+            f'Initial pipeline was fitted in {round(self.timer.assumption_fit_spend_time.total_seconds(), 1)} sec.')
 
         n_jobs = determine_n_jobs(api_params['n_jobs'])
         self.preset_name = assumption_handler.propose_preset(preset, self.timer, n_jobs=n_jobs)
@@ -302,8 +302,8 @@ def tune_final_pipeline(self, task: Task,
         if self.timer.have_time_for_tuning():
             # Tune all nodes in the pipeline
             with self.timer.launch_tuning():
-                self.log.message(f'Hyperparameters tuning started with {round(timeout_for_tuning)} sec. timeout')
                 self.was_tuned = False
+                self.log.message(f'Hyperparameters tuning started with {round(timeout_for_tuning)} min. timeout')
                 tuned_pipeline = tuner.tune(pipeline_gp_composed)
                 self.was_tuned = True
                 self.log.message('Hyperparameters tuning finished')

fedot/api/api_utils/params.py

@@ -118,7 +118,7 @@ def get_default_evo_params(problem: str):
         params = {'max_depth': 6,
                   'max_arity': 3,
                   'pop_size': 20,
-                  'num_of_generations': 100,
+                  'num_of_generations': None,
                   'keep_n_best': 1,
                   'with_tuning': True,
                   'preset': AUTO_PRESET_NAME,

fedot/core/constants.py

@@ -1,7 +1,7 @@
 from fedot.core.repository.tasks import TaskTypesEnum
 
 MINIMAL_SECONDS_FOR_TUNING = 15
-DEFAULT_TUNING_ITERATIONS_NUMBER = 1000
+DEFAULT_TUNING_ITERATIONS_NUMBER = 100000
 MAXIMAL_ATTEMPTS_NUMBER = 1000
 EVALUATION_ATTEMPTS_NUMBER = 5
 DEFAULT_API_TIMEOUT_MINUTES = 5.0

fedot/core/optimisers/composer_requirements.py

@@ -26,7 +26,7 @@ class ComposerRequirements:
     :param validation_blocks: number of validation blocks for time series validation
     """
 
-    num_of_generations: int = 20
+    num_of_generations: Optional[int] = None
     timeout: Optional[datetime.timedelta] = datetime.timedelta(minutes=5)
     early_stopping_generations: Optional[int] = 10
 

fedot/core/optimisers/gp_comp/gp_optimizer.py

@@ -62,7 +62,8 @@ def __init__(self,
     def _initial_population(self, evaluator: Callable):
         """ Initializes the initial population """
         # Adding of initial assumptions to history as zero generation
-        self._update_population(evaluator(self.initial_individuals))
+        evaluation_result = evaluator(self.initial_individuals)
+        self._update_population(evaluation_result)
 
         if len(self.initial_individuals) < self.graph_optimizer_params.pop_size:
             self.initial_individuals = self._extend_population(self.initial_individuals)

fedot/core/optimisers/populational_optimizer.py

@@ -54,7 +54,7 @@ def __init__(self,
         # early_stopping_generations may be None, so use some obvious max number
         max_stagnation_length = requirements.early_stopping_generations or requirements.num_of_generations
         self.stop_optimization = \
-            GroupedCondition().add_condition(
+            GroupedCondition(results_as_message=True).add_condition(
                 lambda: self.timer.is_time_limit_reached(self.current_generation_num),
                 'Optimisation stopped: Time limit is reached'
             ).add_condition(

fedot/core/pipelines/tuning/tuner_builder.py

@@ -1,5 +1,5 @@
 from datetime import timedelta
-from typing import Callable, ClassVar, Type
+from typing import Callable, Type, Union
 
 from hyperopt import tpe
 
@@ -66,7 +66,7 @@ def with_timeout(self, timeout: timedelta):
         self.timeout = timeout
         return self
 
-    def with_eval_time_constraint(self, eval_time_constraint: timedelta):
+    def with_eval_time_constraint(self, eval_time_constraint: Union[timedelta, int, float]):
         self.eval_time_constraint = eval_time_constraint
         return self
 

fedot/core/utilities/grouped_condition.py

@@ -14,10 +14,11 @@ class GroupedCondition:
     if any of the composed conditions is True. The message corresponding
     to the actual fired condition is logged (if it was provided)."""
 
-    def __init__(self, conditions_reduce: Callable[[Iterable[bool]], bool] = any):
+    def __init__(self, conditions_reduce: Callable[[Iterable[bool]], bool] = any, results_as_message: bool = False):
         self._reduce = conditions_reduce
         self._conditions: List[ConditionEntryType] = []
         self._log = default_log(self)
+        self._results_as_message = results_as_message
 
     def add_condition(self, condition: ConditionType, log_msg: Optional[str] = None) -> 'GroupedCondition':
         """Builder-like method for adding conditions."""
@@ -34,5 +35,8 @@ def _check_condition(self, entry: ConditionEntryType) -> bool:
         cond, msg = entry
         res = cond()
         if res and msg:
-            self._log.info(msg)
+            if self._results_as_message:
+                self._log.message(msg)
+            else:
+                self._log.info(msg)
         return res

requirements.txt

@@ -1,5 +1,5 @@
 # Data
-numpy>=1.21.*
+numpy>=1.16.*
 pandas>=1.1.0, <1.3.0; python_version == '3.7'
 pandas>=1.3.0; python_version >='3.8'
 
@@ -13,10 +13,10 @@ ete3>=3.1.*
 networkx>=2.4, !=2.7.*, !=2.8.1, !=2.8.2, !=2.8.3
 scikit_learn
 scikit_learn>=1.0.0; python_version == '3.7'
-scikit_learn>=1.1.0; python_version >= '3.8'
+scikit_learn>=1.0.0; python_version >= '3.8'
 
 # Analysis and optimizations
-hyperopt==0.2.5
+hyperopt==0.2.7
 SALib>=1.3.*
 scikit-optimize>=0.7.4