Adds regression tests for #2007

tests/ludwig/features/test_audio_feature.py

@@ -1,11 +1,15 @@
+import os
 from random import choice
 from string import ascii_lowercase, ascii_uppercase, digits
-from typing import Dict
 
+import pandas as pd
 import pytest
 import torch
 
-from ludwig.features.audio_feature import AudioInputFeature
+from ludwig.backend import LOCAL_BACKEND
+from ludwig.constants import FILL_WITH_MEAN, PROC_COLUMN
+from ludwig.features.audio_feature import AudioFeatureMixin, AudioInputFeature
+from tests.integration_tests.utils import audio_feature, category_feature, generate_data
 
 BATCH_SIZE = 2
 SEQ_SIZE = 20
@@ -16,9 +20,9 @@
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
 
-@pytest.fixture(scope="module")
-def audio_config():
-    return {
+@pytest.mark.parametrize("encoder", ["rnn", "stacked_cnn", "parallel_cnn", "stacked_parallel_cnn", "rnn", "cnnrnn"])
+def test_audio_input_feature(encoder: str) -> None:
+    audio_config = {
         "name": "audio_feature",
         "type": "audio",
         "preprocessing": {
@@ -34,13 +38,53 @@ def audio_config():
         "vocab": VOCAB,
         "max_sequence_length": SEQ_SIZE,
         "embedding_size": AUDIO_W_SIZE,
+        "encoder": encoder,
     }
-
-
-@pytest.mark.parametrize("encoder", ["rnn", "stacked_cnn", "parallel_cnn", "stacked_parallel_cnn", "rnn", "cnnrnn"])
-def test_audio_input_feature(audio_config: Dict, encoder: str) -> None:
-    audio_config.update({"encoder": encoder})
     audio_input_feature = AudioInputFeature(audio_config).to(DEVICE)
     audio_tensor = torch.randn([BATCH_SIZE, SEQ_SIZE, AUDIO_W_SIZE], dtype=torch.float32).to(DEVICE)
     encoder_output = audio_input_feature(audio_tensor)
     assert encoder_output["encoder_output"].shape[1:] == audio_input_feature.output_shape
+
+
+@pytest.mark.parametrize("feature_type", ["raw", "stft", "stft_phase", "group_delay", "fbank"])
+def test_add_feature_data(feature_type, tmpdir):
+    preprocessing_params = {
+        "audio_file_length_limit_in_s": 3.0,
+        "missing_value_strategy": FILL_WITH_MEAN,
+        "in_memory": True,
+        "padding_value": 0,
+        "norm": "per_file",
+        "audio_feature": {
+            "type": feature_type,
+            "window_length_in_s": 0.04,
+            "window_shift_in_s": 0.02,
+            "num_filter_bands": 80,
+        },
+    }
+    audio_dest_folder = os.path.join(tmpdir, "generated_audio")
+    audio_feature_config = audio_feature(audio_dest_folder, preprocessing=preprocessing_params)
+    data_df_path = generate_data(
+        [audio_feature_config],
+        [category_feature(vocab_size=5, reduce_input="sum")],
+        os.path.join(tmpdir, "data.csv"),
+        num_examples=10,
+    )
+    data_df = pd.read_csv(data_df_path)
+    metadata = {
+        audio_feature_config["name"]: AudioFeatureMixin.get_feature_meta(
+            data_df[audio_feature_config["name"]], preprocessing_params, LOCAL_BACKEND
+        )
+    }
+
+    proc_df = {}
+    AudioFeatureMixin.add_feature_data(
+        feature_config=audio_feature_config,
+        input_df=data_df,
+        proc_df=proc_df,
+        metadata=metadata,
+        preprocessing_parameters=preprocessing_params,
+        backend=LOCAL_BACKEND,
+        skip_save_processed_input=False,
+    )
+
+    assert len(proc_df[audio_feature_config[PROC_COLUMN]]) == 10