Training test cases

.github/workflows/ci.yml

@@ -37,4 +37,4 @@ jobs:
         run: |
           python -m pip install .[tests]
           python -m birdnet_analyzer.utils
-          python -m pytest
+          python -m pytest -s tests/test_utils.py tests/train/

birdnet_analyzer/audio.py

@@ -27,6 +27,7 @@ def open_audio_file(path: str, sample_rate=48000, offset=0.0, duration=None, fmi
     Returns:
         Returns the audio time series and the sampling rate.
     """
+
     # Open file with librosa (uses ffmpeg or libav)
     if speed == 1.0:
         sig, rate = librosa.load(

birdnet_analyzer/model.py

@@ -1197,7 +1197,6 @@ def embeddings(sample):
     Returns:
         The embeddings.
     """
-
     load_model(False)
 
     sample = np.array(sample, dtype="float32")

birdnet_analyzer/train/utils.py

@@ -88,6 +88,7 @@ def _load_audio_file(f, label_vector, config):
     else:
         sig_splits = audio.split_signal(sig, rate, cfg.SIG_LENGTH, cfg.SIG_OVERLAP, cfg.SIG_MINLEN)
 
+
     # Get feature embeddings
     batch_size = 1  # turns out that batch size 1 is the fastest, probably because of having to resize the model input when the number of samples in a batch changes
     for i in range(0, len(sig_splits), batch_size):

tests/embeddings/test_embeddings.py

@@ -4,9 +4,11 @@
 import tempfile
 from unittest.mock import MagicMock, patch
 
+import numpy as np
 import pytest
 
 import birdnet_analyzer.config as cfg
+from birdnet_analyzer import model
 from birdnet_analyzer.cli import embeddings_parser
 from birdnet_analyzer.embeddings.core import embeddings
 
@@ -53,3 +55,16 @@ def test_embeddings_cli(mock_run_embeddings: MagicMock, mock_ensure_model: Magic
     mock_ensure_model.assert_called_once()
     threads = min(8, max(1, multiprocessing.cpu_count() // 2))
     mock_run_embeddings.assert_called_once_with(env["input_dir"], env["output_dir"], 0, 1.0, 0, 15000, threads, 1, None)
+
+
+def test_model_embeddings_function_returns_expected_shape():
+    # Create a dummy sample (e.g., 1D numpy array of audio data)
+    sample = np.zeros(144000).astype(np.float32)
+    # Reshape the sample to (1, 144000) as expected by the model
+    sample = sample.reshape(1, 144000)
+    # Call the embeddings function
+    result = model.embeddings(sample)
+
+    # Check that result is a numpy array and has expected shape (depends on model, e.g., (1, embedding_dim))
+    assert isinstance(result, np.ndarray)
+    assert result.ndim == 2

tests/train/test_train.py

@@ -6,6 +6,7 @@
 import pytest
 
 import birdnet_analyzer.config as cfg
+from birdnet_analyzer.analyze.core import analyze
 from birdnet_analyzer.cli import train_parser
 from birdnet_analyzer.train.core import train
 
@@ -17,10 +18,11 @@ def setup_test_environment():
     input_dir = os.path.join(test_dir, "input")
     output_dir = os.path.join(test_dir, "output")
 
-    os.makedirs(input_dir, exist_ok=True)
-    os.makedirs(output_dir, exist_ok=True)
+    # Directory should not exist, so no exist_ok=True
+    os.makedirs(input_dir)
+    os.makedirs(output_dir)
 
-    classifier_output = os.path.join(output_dir, "classifier_output")
+    classifier_output = os.path.join(output_dir, "classifier_output", "custom_classifier.tflite")
 
     # Store original config values
     original_config = {
@@ -55,3 +57,32 @@ def test_train_cli(mock_train_model, mock_ensure_model, setup_test_environment):
 
     mock_ensure_model.assert_called_once()
     mock_train_model.assert_called_once_with()
+
+@pytest.mark.timeout(400)  # Increase timeout for training, 400s should be sufficient, win is by far the slowest
+def test_training(setup_test_environment):
+    """Test the training process and prediction with dummy data."""
+    env = setup_test_environment
+    training_data_input = "tests/data/training"
+
+    # Read class names from subfolders in the input directory, filtering out background classes
+    dummy_classes = [
+        d for d in os.listdir(training_data_input)
+        if os.path.isdir(os.path.join(training_data_input, d)) and d.lower() not in cfg.NON_EVENT_CLASSES
+    ]
+
+    train(training_data_input, env["classifier_output"])
+
+    assert os.path.isfile(env["classifier_output"]), "Classifier output file was not created."
+    assert os.path.exists(env["classifier_output"].replace(".tflite", "_Labels.txt")), "Labels file was not created."
+    assert os.path.exists(env["classifier_output"].replace(".tflite", "_Params.csv")), "Params file was not created."
+    assert os.path.exists(env["classifier_output"].replace(".tflite", ".tflite_sample_counts.csv")), "Params file was not created."
+
+    soundscape_path = "birdnet_analyzer/example/soundscape.wav"
+    analyze(soundscape_path, env["output_dir"], top_n=1, classifier=env["classifier_output"])
+
+    output_file = os.path.join(env["output_dir"], "soundscape.BirdNET.selection.table.txt")
+    with open(output_file) as f:
+        lines = f.readlines()[1:]
+        for line in lines:
+            parts = line.strip().split("\t")
+            assert parts[7] in dummy_classes, f"Detected class {parts[7]} not in expected classes {dummy_classes}"