Bug with zero TS

[aPovidlo]

Fedot crushing after fitting it with ts full of zero values.

Code to reproduce:

horizon = 10
task_params = TsForecastingParams(forecast_length=horizon)
task = Task(TaskTypesEnum.ts_forecasting, task_params)

input_data = InputData(
    idx=np.arange(0, 50),
    features=np.zeros(50),
    target=np.zeros(50),
    task=task,
    data_type=DataTypesEnum.ts
)

auto_model = Fedot(
    problem='ts_forecasting',
    task_params=task_params,
    timeout=4,
    n_jobs=-1,
    seed=15
)

pipeline = auto_model.fit(input_data)

Traceback^

Generations:   0%|          | 1/10000 [02:24<?, ?gen/s]
Traceback (most recent call last):
  File "C:\Users\andre\AppData\Local\Programs\Python\Python39\lib\contextlib.py", line 135, in __exit__
    self.gen.throw(type, value, traceback)
  File "C:\Users\andre\Documents\GitHub\FEDOT\fedot\api\time.py", line 58, in launch_composing
    yield
  File "C:\Users\andre\Documents\GitHub\FEDOT\fedot\api\api_utils\api_composer.py", line 130, in compose_pipeline
    best_pipelines = gp_composer.compose_pipeline(data=train_data)
  File "C:\Users\andre\Documents\GitHub\FEDOT\fedot\core\composer\gp_composer\gp_composer.py", line 67, in compose_pipeline
    opt_result = self.optimizer.optimise(objective_function)
  File "C:\Users\andre\Documents\GitHub\FEDOT\venv\lib\site-packages\golem\core\optimisers\populational_optimizer.py", line 98, in optimise
    new_population = self._evolve_population(evaluator)
  File "C:\Users\andre\Documents\GitHub\FEDOT\venv\lib\site-packages\golem\core\optimisers\genetic\gp_optimizer.py", line 112, in _evolve_population
    new_population = self.reproducer.reproduce(individuals_to_select, evaluator)
  File "C:\Users\andre\Documents\GitHub\FEDOT\venv\lib\site-packages\golem\core\optimisers\genetic\operators\reproduction.py", line 106, in reproduce
    collected_next_population.update({ind.uid: ind for ind in partial_next_population})
TypeError: 'NoneType' object is not iterable

Sometimes, it cause Exception

Exception                                 Traceback (most recent call last)
Cell In[30], line 8
      4 locs.remove(base_location)
      6 print(f'Base TS was choosen randomly: {base_location}')
----> 8 model, pipeline, forecast = fit_predict_model(base_location, plots=True)
     10 test['emission'][test['location'] == base_location] = forecast
     12 for loc in tqdm(locs):

Cell In[28], line 12, in fit_predict_model(loc_name, plots)
      2 ts_train, ts_test, train_data, _, task_parameters = prepare_data(loc_name)
      4 auto_model = Fedot(
      5     problem='ts_forecasting',
      6     task_params=task_parameters,
   (...)
      9     seed=15,
     10 )
---> 12 pipeline = auto_model.fit(features=train_data)
     14 pipeline.show(node_size_scale=0.5, dpi=100)
     16 if plots:

File /opt/conda/lib/python3.10/site-packages/fedot/api/main.py:247, in Fedot.fit(self, features, target, predefined_model)
    243     self.current_pipeline = PredefinedModel(predefined_model, self.train_data, self.log,
    244                                             use_input_preprocessing=self.params.get(
    245                                                 'use_input_preprocessing')).fit()
    246 else:
--> 247     self.current_pipeline, self.best_models, self.history = self.api_composer.obtain_model(self.train_data)
    249     if self.current_pipeline is None:
    250         raise ValueError('No models were found')

File /opt/conda/lib/python3.10/site-packages/fedot/api/api_utils/api_composer.py:69, in ApiComposer.obtain_model(self, train_data)
     62 self.params.init_params_for_composing(self.timer.timedelta_composing, multi_objective)
     64 self.log.message(f"AutoML configured."
     65                  f" Parameters tuning: {with_tuning}."
     66                  f" Time limit: {timeout} min."
     67                  f" Set of candidate models: {self.params.get('available_operations')}.")
---> 69 best_pipeline, best_pipeline_candidates, gp_composer = self.compose_pipeline(
     70     train_data,
     71     initial_assumption,
     72     fitted_assumption
     73 )
     74 if with_tuning:
     75     best_pipeline = self.tune_final_pipeline(train_data, best_pipeline)

File /opt/conda/lib/python3.10/site-packages/fedot/api/api_utils/api_composer.py:130, in ApiComposer.compose_pipeline(self, train_data, initial_assumption, fitted_assumption)
    128 self.log.message('Pipeline composition started.')
    129 self.was_optimised = False
--> 130 best_pipelines = gp_composer.compose_pipeline(data=train_data)
    131 best_pipeline_candidates = gp_composer.best_models
    132 self.was_optimised = True

File /opt/conda/lib/python3.10/site-packages/fedot/core/composer/gp_composer/gp_composer.py:67, in GPComposer.compose_pipeline(self, data)
     64     self.optimizer.set_evaluation_callback(objective_evaluator.evaluate_intermediate_metrics)
     66 # Finally, run optimization process
---> 67 opt_result = self.optimizer.optimise(objective_function)
     69 best_model, self.best_models = self._convert_opt_results_to_pipeline(opt_result)
     70 self.log.info('GP composition finished')

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/populational_optimizer.py:98, in PopulationalOptimizer.optimise(self, objective)
     96 while not self.stop_optimization():
     97     try:
---> 98         new_population = self._evolve_population(evaluator)
     99         if self.gen_structural_diversity_check != -1 \
    100                 and self.generations.generation_num % self.gen_structural_diversity_check == 0 \
    101                 and self.generations.generation_num != 0:
    102             new_population = self.get_structure_unique_population(new_population, evaluator)

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/genetic/gp_optimizer.py:112, in EvoGraphOptimizer._evolve_population(self, evaluator)
    110 individuals_to_select = self.regularization(self.population, evaluator)
    111 # Reproduce from previous pop to get next population
--> 112 new_population = self.reproducer.reproduce(individuals_to_select, evaluator)
    114 # Adaptive agent experience collection & learning
    115 # Must be called after reproduction (that collects the new experience)
    116 experience = self.mutation.agent_experience

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/genetic/operators/reproduction.py:104, in ReproductionController.reproduce(self, population, evaluator)
    101 residual_size = min(len(population), residual_size)
    103 # Reproduce the required number of individuals that equals residual size
--> 104 partial_next_population = self.reproduce_uncontrolled(population, evaluator, residual_size)
    105 # Avoid duplicate individuals that can come unchanged from previous population
    106 collected_next_population.update({ind.uid: ind for ind in partial_next_population})

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/genetic/operators/reproduction.py:82, in ReproductionController.reproduce_uncontrolled(self, population, evaluator, pop_size)
     80 new_population = self.crossover(selected_individuals)
     81 new_population = self.mutation(new_population)
---> 82 new_population = evaluator(new_population)
     83 return new_population

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/genetic/evaluation.py:199, in BaseGraphEvaluationDispatcher.evaluate_with_cache(self, population)
    197 reversed_population = list(reversed(population))
    198 self._remote_compute_cache(reversed_population)
--> 199 evaluated_population = self.evaluate_population(reversed_population)
    200 self._reset_eval_cache()
    201 return evaluated_population

File /opt/conda/lib/python3.10/site-packages/golem/core/optimisers/genetic/evaluation.py:249, in MultiprocessingDispatcher.evaluate_population(self, individuals)
    247 parallel = Parallel(n_jobs=n_jobs, verbose=0, pre_dispatch="2*n_jobs")
    248 eval_func = partial(self.evaluate_single, logs_initializer=Log().get_parameters())
--> 249 evaluation_results = parallel(delayed(eval_func)(ind.graph, ind.uid) for ind in individuals_to_evaluate)
    250 individuals_evaluated = self.apply_evaluation_results(individuals_to_evaluate, evaluation_results)
    251 # If there were no successful evals then try once again getting at least one,
    252 # even if time limit was reached

File /opt/conda/lib/python3.10/site-packages/joblib/parallel.py:1098, in Parallel.__call__(self, iterable)
   1095     self._iterating = False
   1097 with self._backend.retrieval_context():
-> 1098     self.retrieve()
   1099 # Make sure that we get a last message telling us we are done
   1100 elapsed_time = time.time() - self._start_time

File /opt/conda/lib/python3.10/site-packages/joblib/parallel.py:975, in Parallel.retrieve(self)
    973 try:
    974     if getattr(self._backend, 'supports_timeout', False):
--> 975         self._output.extend(job.get(timeout=self.timeout))
    976     else:
    977         self._output.extend(job.get())

File /opt/conda/lib/python3.10/site-packages/joblib/_parallel_backends.py:567, in LokyBackend.wrap_future_result(future, timeout)
    564 """Wrapper for Future.result to implement the same behaviour as
    565 AsyncResults.get from multiprocessing."""
    566 try:
--> 567     return future.result(timeout=timeout)
    568 except CfTimeoutError as e:
    569     raise TimeoutError from e

File /opt/conda/lib/python3.10/concurrent/futures/_base.py:458, in Future.result(self, timeout)
    456     raise CancelledError()
    457 elif self._state == FINISHED:
--> 458     return self.__get_result()
    459 else:
    460     raise TimeoutError()

File /opt/conda/lib/python3.10/concurrent/futures/_base.py:403, in Future.__get_result(self)
    401 if self._exception:
    402     try:
--> 403         raise self._exception
    404     finally:
    405         # Break a reference cycle with the exception in self._exception
    406         self = None

Exception: Invalid fitness after objective evaluation. Skipping the graph: (/n_lagged_{'window_size': 10};)/n_cgru_{'hidden_size': 200, 'learning_rate': 0.001, 'cnn1_kernel_size': 5, 'cnn1_output_size': 32, 'cnn2_kernel_size': 5, 'cnn2_output_size': 64, 'batch_size': 64, 'num_epochs': 50, 'optimizer': 'adamw', 'loss': 'mse'}

[nicl-nno]

#1095 может быть связано с той же проблемой.