Added docker and k8s stuff

cnn_classification_service/cnn_classifier.py

@@ -0,0 +1,87 @@
+from typing import Tuple
+import tempfile
+import os
+import os.path
+import shutil
+
+import librosa
+import librosa.display
+import numpy
+import matplotlib.pyplot
+from keras.models import model_from_json
+from keras import optimizers
+from keras_preprocessing.image import ImageDataGenerator
+
+
+class Classifier(object):
+
+    def __init__(self, model_filename: str, weights_filename: str):
+        with open(model_filename, 'r') as f:
+            self.loaded_model = model_from_json(f.read())
+
+        self.loaded_model.load_weights(weights_filename)
+        self.datagen = ImageDataGenerator(rescale=1. / 255., validation_split=0.25)
+        self.loaded_model.compile(optimizers.rmsprop(lr=0.0005, decay=1e-6), loss="categorical_crossentropy",
+                                  metrics=["accuracy"])
+        self.loaded_model.summary()
+
+    @staticmethod
+    def create_spectrogram(wav_filename: str) -> Tuple[str, str]:
+        matplotlib.pyplot.interactive(False)
+        clip, sample_rate = librosa.load(wav_filename, sr=None)
+        fig = matplotlib.pyplot.figure(figsize=[0.72, 0.72])
+        ax = fig.add_subplot(111)
+        ax.axes.get_xaxis().set_visible(False)
+        ax.axes.get_yaxis().set_visible(False)
+        ax.set_frame_on(False)
+        spectogram = librosa.feature.melspectrogram(y=clip, sr=sample_rate)
+        librosa.display.specshow(librosa.power_to_db(spectogram, ref=numpy.max))
+
+        target_dir = tempfile.mkdtemp()
+
+        # Change extension to jpg... mert 110% biztos vagyok benne hogy a keras nem bírná beolvasni máshogy
+        file_name = os.path.join(target_dir, "unknown", f"{wav_filename[:-4]}.jpg")
+
+        matplotlib.pyplot.savefig(file_name, dpi=400, bbox_inches='tight', pad_inches=0)
+        matplotlib.pyplot.close()
+        fig.clf()
+        matplotlib.pyplot.close(fig)
+        matplotlib.pyplot.close('all')
+
+        return target_dir, file_name
+
+    def _run_predictor(self, directory: str) -> list:
+        predict_generator = self.datagen.flow_from_directory(
+            directory=directory,
+            batch_size=128,
+            seed=42,
+            shuffle=False,
+            class_mode="categorical",
+            target_size=(64, 64))
+
+        prediction = self.loaded_model.predict_generator(predict_generator, steps=1)
+
+        predicted_class_indices = numpy.argmax(prediction, axis=1)
+
+        labels = {
+            'anser': 0,
+            'columba': 1,
+            'hirundo': 2,
+            'passer': 3,
+            'sturnus': 4,
+            'turdus': 5,
+            'upupa': 6
+        }
+        labels = dict((v, k) for k, v in labels.items())
+
+        predictions = [labels[k] for k in predicted_class_indices]
+
+        return predictions
+
+    def predict(self, wav_filename: str) -> list:
+        directory, _ = self.create_spectrogram(wav_filename)
+
+        result = self._run_predictor(directory)
+        shutil.rmtree(directory)  # The image is no longer needed
+
+        return result