- added pipeline tuning to multi_modal_genre_prediction.py

- now there is no useless try of download of stopwords and other nltk packages if they are already downloaded
- keras.Input changed to recommended keras.layers.InputLayer
- test_multi_modal.py is moved to multimodal folder

cases/multi_modal_genre_prediction.py

@@ -1,4 +1,7 @@
 import datetime
+
+from sklearn.metrics import f1_score
+
 from examples.advanced.multi_modal_pipeline import calculate_validation_metric, \
     generate_initial_pipeline_and_data, prepare_multi_modal_data
 from fedot.core.composer.composer_builder import ComposerBuilder
@@ -7,6 +10,7 @@
 from fedot.core.optimisers.gp_comp.gp_optimiser import GPGraphOptimiserParameters, GeneticSchemeTypesEnum
 from fedot.core.repository.operation_types_repository import get_operations_for_task
 from fedot.core.repository.quality_metrics_repository import ClassificationMetricsEnum
+from fedot.core.pipelines.tuning.unified import PipelineTuner
 from fedot.core.repository.tasks import Task, TaskTypesEnum
 
 
@@ -49,13 +53,20 @@ def run_multi_modal_case(files_path, is_visualise=True, timeout=datetime.timedel
     # the optimal pipeline generation by composition - the most time-consuming task
     pipeline_evo_composed = composer.compose_pipeline(data=fit_data,
                                                       is_visualise=True)
+    pipeline_evo_composed.print_structure()
 
-    pipeline_evo_composed.fit(input_data=fit_data)
+    # tuning of the composed pipeline
+    pipeline_tuner = PipelineTuner(pipeline=pipeline_evo_composed, task=task, iterations=15)
+    tuned_pipeline = pipeline_tuner.tune_pipeline(input_data=fit_data,
+                                                  loss_function=f1_score,
+                                                  loss_params={'average': 'micro'})
+    tuned_pipeline.print_structure()
+    tuned_pipeline.fit(input_data=fit_data)
 
     if is_visualise:
-        pipeline_evo_composed.show()
+        tuned_pipeline.show()
 
-    prediction = pipeline_evo_composed.predict(predict_data, output_mode='labels')
+    prediction = tuned_pipeline.predict(predict_data, output_mode='labels')
     err = calculate_validation_metric(predict_data, prediction)
 
     print(f'F1 micro for validation sample is {err}')

fedot/core/operations/evaluation/operation_implementations/data_operations/text_preprocessing.py

@@ -61,11 +61,16 @@ def transform(self, input_data, is_fit_pipeline_stage: Optional[bool]):
 
     @staticmethod
     def _download_nltk_resources():
-        for resource in ['punkt', 'stopwords', 'wordnet', 'omw-1.4']:
+        for resource in ['punkt']:
             try:
                 nltk.data.find(f'tokenizers/{resource}')
             except LookupError:
                 nltk.download(f'{resource}')
+        for resource in ['stopwords', 'wordnet', 'omw-1.4']:
+            try:
+                nltk.data.find(f'corpora/{resource}')
+            except LookupError:
+                nltk.download(f'{resource}')
 
     @staticmethod
     def _word_vectorize(text):

fedot/core/operations/evaluation/operation_implementations/models/keras.py

@@ -32,7 +32,7 @@ def create_deep_cnn(input_shape: tuple,
                     num_classes: int):
     model = tf.keras.Sequential(
         [
-            tf.keras.Input(shape=input_shape),
+            tf.keras.layers.InputLayer(input_shape=input_shape),
             tf.keras.layers.Conv2D(32, kernel_size=(3, 3), activation="relu"),
             tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),
             tf.keras.layers.Conv2D(64, kernel_size=(3, 3), activation="relu"),
@@ -51,7 +51,7 @@ def create_simple_cnn(input_shape: tuple,
                       num_classes: int):
     model = tf.keras.Sequential(
         [
-            tf.keras.Input(shape=input_shape),
+            tf.keras.layers.InputLayer(input_shape=input_shape),
             tf.keras.layers.Conv2D(32, kernel_size=(3, 3), activation="relu"),
             tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),
             tf.keras.layers.Flatten(),

test/unit/pipelines/test_multi_modal.py → test/unit/multimodal/test_multi_modal.py

@@ -1,7 +1,6 @@
 import os
 
 from examples.advanced.multi_modal_pipeline import (prepare_multi_modal_data)
-from fedot.core.data.multi_modal import MultiModalData
 from fedot.core.pipelines.node import PrimaryNode, SecondaryNode
 from fedot.core.pipelines.pipeline import Pipeline
 from fedot.core.repository.tasks import Task, TaskTypesEnum