{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"## Imports","metadata":{}},{"cell_type":"code","source":"import os\nimport json \nimport librosa\nimport numpy as np\nimport pandas as pd\nimport soundfile as sf\nfrom sklearn.model_selection import train_test_split\nimport tensorflow as tf\nimport os\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import GlobalAveragePooling2D, GlobalMaxPooling2D, Dense,Conv2D,Flatten,MaxPooling2D,Dropout, Activation, BatchNormalization\nfrom tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau\nfrom tensorflow.keras.optimizers import Adam, SGD\nfrom tensorflow.keras import optimizers\nfrom tensorflow.keras.utils import plot_model, Sequence\nfrom tensorflow.keras.utils import to_categorical\nfrom tensorflow.keras.applications import MobileNetV2\nfrom tensorflow.keras.layers import Input, Concatenate\nimport tensorflow as tf\nfrom tensorflow import keras\nfrom tensorflow.keras import layers","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2022-06-13T16:31:48.745911Z","iopub.execute_input":"2022-06-13T16:31:48.746382Z","iopub.status.idle":"2022-06-13T16:31:55.675001Z","shell.execute_reply.started":"2022-06-13T16:31:48.746272Z","shell.execute_reply":"2022-06-13T16:31:55.674231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Adding data","metadata":{}},{"cell_type":"code","source":"train_meta = pd.read_csv(\"/kaggle/input/birdclef-2022/train_metadata.csv\")\nlabels = list(train_meta['primary_label'].unique())\ntrain_meta","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:31:55.677545Z","iopub.execute_input":"2022-06-13T16:31:55.678045Z","iopub.status.idle":"2022-06-13T16:31:55.814536Z","shell.execute_reply.started":"2022-06-13T16:31:55.678007Z","shell.execute_reply":"2022-06-13T16:31:55.813798Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Pre-processing ","metadata":{}},{"cell_type":"code","source":"\ndef cutAudio(file_path, is_save):\n    # First load the file\n    filename = file_path.replace(\"/\", \"_\")\n    file_path = \"/kaggle/input/birdclef-2022/train_audio/\" + file_path\n    audio, sr = librosa.load(file_path)\n    \n    # Get number of samples for 5 seconds; replace 5 by any number\n    buffer = 5 * sr\n\n    samples_total = len(audio)\n    samples_wrote = 0\n    counter = 1\n\n    audio_split = []\n    audio_filenames = []\n    while samples_wrote < samples_total:\n        #check if the buffer is not exceeding total samples \n        if buffer > (samples_total - samples_wrote):\n            buffer = samples_total - samples_wrote\n\n        block = audio[samples_wrote : (samples_wrote + buffer)]\n        audio_split.append(block)\n\n        # Write 5 second segment\n        if is_save == True:\n            out_filename = \"/kaggle/working/each5s/split_\" + str(counter) + \"_\" + filename\n            audio_filenames.append(out_filename)\n            sf.write(out_filename, block, sr)\n        counter += 1\n        samples_wrote += buffer\n    return audio_split, sr, audio_filenames","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:31:55.815748Z","iopub.execute_input":"2022-06-13T16:31:55.817112Z","iopub.status.idle":"2022-06-13T16:31:55.825046Z","shell.execute_reply.started":"2022-06-13T16:31:55.817073Z","shell.execute_reply":"2022-06-13T16:31:55.823958Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def splitTrainAudio(_df):\n    data = []\n    for index, row in _df.iterrows():\n        cutAudio(row[\"filename\"], True)\n        audio_lst, sr, filenames = cutAudio(row[\"filename\"], True)\n        for idx, y in enumerate(audio_lst):\n            data.append([row[\"primary_label\"], row[\"filename\"], filenames[idx]])\n\n    data_df = pd.DataFrame(data, columns=['primary_label', 'original_filename', 'filename'])\n    data_df.to_csv(\"/kaggle/working/data_df.csv\", index=False)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:31:55.82675Z","iopub.execute_input":"2022-06-13T16:31:55.827136Z","iopub.status.idle":"2022-06-13T16:31:55.836496Z","shell.execute_reply.started":"2022-06-13T16:31:55.827092Z","shell.execute_reply":"2022-06-13T16:31:55.835815Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Sample Data\ndata_frames = []\nfor label in labels:\n    tmp_df = train_meta[train_meta[\"primary_label\"] == label].sample(n=2, replace=True).reset_index(drop=True)\n    data_frames.append(tmp_df)\nsample_df = pd.concat(data_frames).reset_index(drop=True)\nsample_df","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:31:55.839787Z","iopub.execute_input":"2022-06-13T16:31:55.840055Z","iopub.status.idle":"2022-06-13T16:31:56.329563Z","shell.execute_reply.started":"2022-06-13T16:31:55.840029Z","shell.execute_reply":"2022-06-13T16:31:56.328837Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!mkdir -p \"/kaggle/working/each5s\"\nsplitTrainAudio(sample_df)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:31:56.330878Z","iopub.execute_input":"2022-06-13T16:31:56.331148Z","iopub.status.idle":"2022-06-13T16:34:22.234698Z","shell.execute_reply.started":"2022-06-13T16:31:56.331113Z","shell.execute_reply":"2022-06-13T16:34:22.233008Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_df = pd.read_csv(\"/kaggle/working/data_df.csv\")\ndata_df","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.236146Z","iopub.status.idle":"2022-06-13T16:34:22.236558Z","shell.execute_reply.started":"2022-06-13T16:34:22.236337Z","shell.execute_reply":"2022-06-13T16:34:22.236359Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Data generation","metadata":{}},{"cell_type":"code","source":"num_rows = 216\nnum_columns = 216\nnum_channels = 1\nn_mels = 512\n\ndef extractFeatures(y, sr):\n    feat = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=num_rows, n_mels=n_mels)\n    if feat.shape[1] <= num_columns:\n        pad_width = num_columns - feat.shape[1]\n        feat = np.pad(feat, pad_width=((0,0),(0,pad_width)), mode='constant')\n    return feat","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.237979Z","iopub.status.idle":"2022-06-13T16:34:22.238375Z","shell.execute_reply.started":"2022-06-13T16:34:22.238162Z","shell.execute_reply":"2022-06-13T16:34:22.238183Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class DataGenerator(Sequence):\n    def __init__(self,\n                _X,\n                batch_size=32,\n                n_channels=1,\n                n_columns=470,\n                n_rows=120,\n                shuffle=True):\n        self.batch_size = batch_size\n        self.X = _X\n        self.n_channels = n_channels\n        self.n_columns = n_columns\n        self.n_rows = n_rows\n        self.shuffle = shuffle\n        self.img_indexes = np.arange(len(self.X))\n        self.on_epoch_end()\n        \n    def __len__(self):\n        'Denotes the number of batches per epoch'\n        return int(np.floor(len(self.img_indexes) / self.batch_size))\n\n    def __getitem__(self, index):\n        'Generate one batch of data'\n        # Generate indexes of the batch\n        indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]\n        # Find list of IDs\n        list_IDs_temps = [self.img_indexes[k] for k in indexes]\n\n        # Generate data\n        X, y = self.__data_generation(list_IDs_temps)\n        return X, y\n\n    def on_epoch_end(self):\n        'Updates indexes after each epoch'\n        self.indexes = np.arange(len(self.X))\n        if self.shuffle == True:\n            np.random.shuffle(self.indexes)\n\n    def __data_generation(self, list_IDs_temps):\n        X = np.empty((self.batch_size, self.n_rows, self.n_columns))\n        y = np.empty((self.batch_size), dtype=int)\n        for i, ID in enumerate(list_IDs_temps):\n            file_path = self.X.iloc[ID][\"filename\"]\n            audio, sr = librosa.load(file_path)       \n            feat = extractFeatures(audio, sr)\n            x_features = feat.tolist()\n            label = self.X.iloc[ID][\"target\"]\n            X[i] = np.array(x_features)\n            y[i] = label\n        X = X.reshape(X.shape[0], self.n_rows, self.n_columns, self.n_channels)\n        \n        return X, to_categorical(y, num_classes=len(labels))\n","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.239888Z","iopub.status.idle":"2022-06-13T16:34:22.240305Z","shell.execute_reply.started":"2022-06-13T16:34:22.240078Z","shell.execute_reply":"2022-06-13T16:34:22.240099Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Plotting results","metadata":{}},{"cell_type":"code","source":"import matplotlib.pyplot as plt\n\ndef plot_his(history):\n    plt.figure(1, figsize = (15,8))\n    plt.subplot(221)\n    plt.plot(history.history['accuracy'])\n    plt.plot(history.history['val_accuracy'])\n    plt.title('model accuracy')\n    plt.ylabel('accuracy')\n    plt.xlabel('epoch')\n    plt.legend(['train', 'valid'])\n    plt.subplot(222)\n    plt.plot(history.history['loss'])\n    plt.plot(history.history['val_loss'])\n    plt.title('model loss')\n    plt.ylabel('loss')\n    plt.xlabel('epoch')\n    plt.legend(['train', 'valid'])\n    plt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.241961Z","iopub.status.idle":"2022-06-13T16:34:22.242552Z","shell.execute_reply.started":"2022-06-13T16:34:22.242326Z","shell.execute_reply":"2022-06-13T16:34:22.24235Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Creating CNN","metadata":{}},{"cell_type":"code","source":"def create_cnn(): \n\n\n    model = Sequential()\n    model.add(layers.Conv2D(32, 3, activation='relu', input_shape=(num_rows, num_columns, num_channels)))\n    model.add(layers.MaxPooling2D((2, 2)))\n    model.add(layers.Dropout(0.2))\n    model.add(layers.Conv2D(64, 3, activation='relu'))\n    model.add(layers.MaxPooling2D((2, 2)))\n    model.add(layers.Dropout(0.2))\n    model.add(layers.Conv2D(64, 3, activation='relu'))\n    model.add(layers.Flatten())\n    model.add(layers.Dense(64, activation='relu'))\n    model.add(layers.Dropout(0.2))\n    model.add(layers.Dense(len(labels), activation='softmax', kernel_regularizer='l1'))\n\n    return model","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.243808Z","iopub.status.idle":"2022-06-13T16:34:22.244225Z","shell.execute_reply.started":"2022-06-13T16:34:22.243999Z","shell.execute_reply":"2022-06-13T16:34:22.244021Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_df['target'] = data_df['primary_label'].apply(lambda x: labels.index(x))\ndata_df","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.245534Z","iopub.status.idle":"2022-06-13T16:34:22.246038Z","shell.execute_reply.started":"2022-06-13T16:34:22.245772Z","shell.execute_reply":"2022-06-13T16:34:22.245797Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Training","metadata":{}},{"cell_type":"code","source":"def train_model(model, train_gen, val_gen):\n    checkpoint_model_path = \"/kaggle/working/mobilnetv2.h5\"\n    metric = \"val_accuracy\"\n    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001), loss='categorical_crossentropy', metrics=[\"accuracy\"])\n    num_epochs = 50\n    checkpointer = ModelCheckpoint(\n        filepath=checkpoint_model_path,\n        monitor=metric, verbose=1, save_best_only=True)\n    es_callback = EarlyStopping(monitor=metric, patience=5, verbose=1)\n    reduce_lr = ReduceLROnPlateau(monitor=metric, factor=0.3, patience=1, verbose=1, min_delta=0.0001, cooldown=1, min_lr=0.00001)\n\n    history = model.fit(\n        train_gen,\n        epochs=num_epochs,\n        validation_data=val_gen,\n        callbacks=[checkpointer,es_callback,reduce_lr],\n        verbose=1\n    )\n\n    plot_his(history)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.24758Z","iopub.status.idle":"2022-06-13T16:34:22.24807Z","shell.execute_reply.started":"2022-06-13T16:34:22.247787Z","shell.execute_reply":"2022-06-13T16:34:22.247813Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"params = dict(\n    batch_size=64,\n    n_rows=num_rows,\n    n_columns=num_columns,\n    n_channels=num_channels,\n)\nparams_train = dict(\n    shuffle=True,\n    **params\n)\nparams_valid = dict(\n    shuffle=False,\n    **params\n)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.249364Z","iopub.status.idle":"2022-06-13T16:34:22.249755Z","shell.execute_reply.started":"2022-06-13T16:34:22.249542Z","shell.execute_reply":"2022-06-13T16:34:22.249564Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_train, X_valid, _, _ = train_test_split(data_df, data_df[\"target\"], test_size=0.1, random_state=1534323, shuffle=True)\ntrain_generator = DataGenerator(X_train, **params_train)\nvalid_generator = DataGenerator(X_valid, **params_valid)\ncnn_model = create_cnn()\ntrain_model(cnn_model, train_generator, valid_generator)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.250848Z","iopub.status.idle":"2022-06-13T16:34:22.252258Z","shell.execute_reply.started":"2022-06-13T16:34:22.252008Z","shell.execute_reply":"2022-06-13T16:34:22.252034Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Submission","metadata":{}},{"cell_type":"code","source":"cnn_model = create_cnn()\ndata_df = pd.read_csv(\"/kaggle/input/birdclef-2022-keras-model/data_df.csv\")\nlabels = list(data_df['primary_label'].unique())","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.253452Z","iopub.status.idle":"2022-06-13T16:34:22.254064Z","shell.execute_reply.started":"2022-06-13T16:34:22.253816Z","shell.execute_reply":"2022-06-13T16:34:22.253839Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_path = \"/kaggle/input/birdclef-2022/test_soundscapes/\"\nfiles = [f.split('.')[0] for f in sorted(os.listdir(test_path))]\n\nbirds_path = \"/kaggle/input/birdclef-2022/scored_birds.json\"\nwith open(birds_path) as bf:\n    birds = json.load(bf)\n\ndata = []\nfor f in files:\n    file_path = test_path + f + '.ogg'\n    audio, sr = librosa.load(file_path)\n    # Get number of samples for 5 seconds; replace 5 by any number\n    buffer = 5 * sr\n    samples_total = len(audio)\n    samples_wrote = 0\n    counter = 1\n\n    while samples_wrote < samples_total:\n        #check if the buffer is not exceeding total samples \n        if buffer > (samples_total - samples_wrote):\n            buffer = samples_total - samples_wrote\n\n        block = audio[samples_wrote : (samples_wrote + buffer)]\n        feat = extractFeatures(block, sr)\n        x = feat.reshape(1, num_rows, num_columns, num_channels)\n        pred = cnn_model.predict(x)\n        label_index = np.argmax(pred,axis=1)[0]\n        \n        for b in birds:\n            segment_end = counter * 5   \n            row_id = f + '_' + b + '_' + str(segment_end)\n            target = False\n            if labels[label_index] == b:\n                target = True\n            data.append([row_id, target])\n        counter += 1\n        samples_wrote += buffer\n        \nsubmission_df = pd.DataFrame(data, columns=['row_id', 'target'])\nsubmission_df","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.255277Z","iopub.status.idle":"2022-06-13T16:34:22.255882Z","shell.execute_reply.started":"2022-06-13T16:34:22.255655Z","shell.execute_reply":"2022-06-13T16:34:22.255679Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission_df.to_csv(\"/kaggle/working/submission.csv\", index=False)","metadata":{"execution":{"iopub.status.busy":"2022-06-13T16:34:22.257093Z","iopub.status.idle":"2022-06-13T16:34:22.257731Z","shell.execute_reply.started":"2022-06-13T16:34:22.257503Z","shell.execute_reply":"2022-06-13T16:34:22.257527Z"},"trusted":true},"execution_count":null,"outputs":[]}]}