Hi everyone, I'm currently trying my best to develop a CNN to classify the birds sound as a beginner data scientist, but I run into an issue while trying to train my model, here is the entire traceback:
\n`---------------------------------------------------------------------------
\nValueError Traceback (most recent call last)
\n/tmp/ipykernel_27/2595944468.py in
\n 48
\n 49 # train model
\n---> 50 train(cnn, train_dataloader, loss_fn, optimiser, device, EPOCHS)
\n 51
\n 52 # save model
/tmp/ipykernel_27/2595944468.py in train(model, data_loader, loss_fn, optimiser, device, epochs)
\n 30 for i in range(epochs):
\n 31 print(f\"Epoch {i+1}\")
\n---> 32 train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n 33 print(\"---------------------------\")
\n 34 print(\"Finished training\")
/tmp/ipykernel_27/2595944468.py in train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n 14 def train_single_epoch(model, data_loader, loss_fn, optimiser, device):
\n 15 for input, target in data_loader:
\n---> 16 input, target = torch.tensor(input).to(device), torch.tensor(target).to(device)
\n 17 # calculate loss
\n 18 prediction = model(input)
ValueError: too many dimensions 'str'`
\nand here is my code:
\n`annotations = pd.read_csv('/kaggle/input/birdclef-2023/train_metadata.csv')
\nannotations.head()
class_names = sorted(os.listdir('/kaggle/input/birdclef-2023/train_audio/'))
\nnum_classes = len(class_names)
\nprint(num_classes)`
`class BirdDataset(Dataset):
\n def init(self, annotations_file, audio_dir, transformation, target_sample_rate, num_samples, device):
\n self.annotations = pd.read_csv(annotations_file)
\n self.audio_dir = audio_dir
\n self.device = device
\n self.transformation = transformation.to(self.device)
\n self.target_sample_rate = target_sample_rate
\n self.num_samples = num_samples
def __len__(self):\n return len(self.annotations)\n\ndef __getitem__(self, index):\n audio_sample_path = self._get_audio_sample_path(index)\n label = self._get_audio_sample_label(index)\n signal, sr = torchaudio.load(audio_sample_path)\n signal = signal.to(self.device)\n signal = self._resample_if_necessary(signal, sr)\n signal = self._mix_down_if_necessary(signal)\n signal = self._right_pad_if_necessary(signal)\n signal = self._cut_if_necessary(signal)\n signal = self.transformation(signal)\n return signal, label\n\ndef _get_audio_sample_path(self, index):\n file_name = self.annotations.iloc[index, 11]\n path = os.path.join(self.audio_dir, file_name)\n return path\n\ndef _get_audio_sample_label(self, index):\n return self.annotations.iloc[index, 0]\n\ndef _resample_if_necessary(self, signal, sr):\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n signal = resample(signal)\n return signal\n\ndef _mix_down_if_necessary(self, signal):\n if signal.shape[0] > 1:\n signal = torch.mean(signal, dim=0, keepdim=True)\n return signal\n\ndef _cut_if_necessary(self, signal):\n if signal.shape[1] > self.num_samples:\n signal = signal[:, :self.num_samples]\n return signal\n\ndef _right_pad_if_necessary(self, signal):\n length_signal = signal.shape[1]\n if length_signal < self.num_samples:\n num_missing_samples = self.num_samples - length_signal\n last_dim_padding = (0, num_missing_sampels)\n signal = torch.nn.functional.pad(signal, last_dim_padding)\n return signal`\n`class CNNNetwork(nn.Module):
\n def init(self):
\n super().init()
self.conv1 = nn.Sequential(\n nn.Conv2d(\n in_channels=1,\n out_channels=16,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n\n self.conv2 = nn.Sequential(\n nn.Conv2d(\n in_channels=16,\n out_channels=32,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.conv3 = nn.Sequential(\n nn.Conv2d(\n in_channels=32,\n out_channels=64,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.conv4 = nn.Sequential(\n nn.Conv2d(\n in_channels=64,\n out_channels=128,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.flatten = nn.Flatten()\n self.linear = nn.Linear(in_features= 128 * 5 * 3 , out_features=num_classes)\n self.softmax = nn.Softmax(dim=1)\n\ndef forward(self, input_data):\n x = self.conv1(input_data)\n x = self.conv2(x)\n x = self.conv3(x)\n x = self.conv4(x)\n x = self.flatten(x)\n logits = self.linear(x)\n predictions = self.softmax(logits)\n return predictions`\n`ANNOTATIONS_FILE = '../input/birdclef-2023/train_metadata.csv'
\nAUDIO_DIR = \"../input/birdclef-2023/train_audio/\"
\nSAMPLE_RATE = 32000
\nNUM_SAMPLES = 32000
device = \"cuda\" if torch.cuda.is_available() else \"cpu\"
\nprint(\"Running on\", device)
mel_spectogram = torchaudio.transforms.MelSpectrogram(sample_rate = SAMPLE_RATE, n_fft=2028, hop_length=1024, n_mels=64)
\nbird_ds = BirdDataset(ANNOTATIONS_FILE, AUDIO_DIR, mel_spectogram, SAMPLE_RATE, NUM_SAMPLES, device)`
`from torch.utils.data import DataLoader
\nBATCH_SIZE = 128
\nEPOCHS = 10
\nLEARNING_RATE = 0.001
device = \"cuda\" if torch.cuda.is_available() else \"cpu\"
\ndef create_data_loader(train_data, batch_size):
\n train_dataloader = DataLoader(train_data, batch_size=batch_size)
\n return train_dataloader
def train_single_epoch(model, data_loader, loss_fn, optimiser, device):
\n for input, target in data_loader:
\n input, target = torch.tensor(input).to(device), torch.tensor(target).to(device)
\n # calculate loss
\n prediction = model(input)
\n loss = loss_fn(prediction, target)
# backpropagate error and update weights\n optimiser.zero_grad()\n loss.backward()\n optimiser.step()\n\nprint(f\"loss: {loss.item()}\")\ndef train(model, data_loader, loss_fn, optimiser, device, epochs):
\n for i in range(epochs):
\n print(f\"Epoch {i+1}\")
\n train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n print(\"---------------------------\")
\n print(\"Finished training\")
train_dataloader = create_data_loader(bird_ds, BATCH_SIZE)
\ncnn = CNNNetwork().to(device)
\nprint(cnn)
loss_fn = nn.CrossEntropyLoss()
\noptimiser = torch.optim.Adam(cnn.parameters(),
\n lr=LEARNING_RATE)
\ntrain(cnn, train_dataloader, loss_fn, optimiser, device, EPOCHS)
\ntorch.save(cnn.state_dict(), \"feedforwardnet.pth\")
\nprint(\"Trained feed forward net saved at feedforwardnet.pth\")`
At first I thought it was extra labels but apparently it's not. I know my code is not optimized at all and the model might create nightmares for advanced people but I'm just trying to learn and get a first prediction to be able to tweak the model after :)
\nI would really appreciate a little boost/help here to what I'm doing wrong :(
\nThank you very much
", "messages": [ { "id": 2229056, "postDate": "2023-04-21T03:56:15.380Z", "content": "Hi everyone, I'm currently trying my best to develop a CNN to classify the birds sound as a beginner data scientist, but I run into an issue while trying to train my model, here is the entire traceback:
\n`---------------------------------------------------------------------------
\nValueError Traceback (most recent call last)
\n/tmp/ipykernel_27/2595944468.py in
\n 48
\n 49 # train model
\n---> 50 train(cnn, train_dataloader, loss_fn, optimiser, device, EPOCHS)
\n 51
\n 52 # save model
/tmp/ipykernel_27/2595944468.py in train(model, data_loader, loss_fn, optimiser, device, epochs)
\n 30 for i in range(epochs):
\n 31 print(f\"Epoch {i+1}\")
\n---> 32 train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n 33 print(\"---------------------------\")
\n 34 print(\"Finished training\")
/tmp/ipykernel_27/2595944468.py in train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n 14 def train_single_epoch(model, data_loader, loss_fn, optimiser, device):
\n 15 for input, target in data_loader:
\n---> 16 input, target = torch.tensor(input).to(device), torch.tensor(target).to(device)
\n 17 # calculate loss
\n 18 prediction = model(input)
ValueError: too many dimensions 'str'`
\nand here is my code:
\n`annotations = pd.read_csv('/kaggle/input/birdclef-2023/train_metadata.csv')
\nannotations.head()
class_names = sorted(os.listdir('/kaggle/input/birdclef-2023/train_audio/'))
\nnum_classes = len(class_names)
\nprint(num_classes)`
`class BirdDataset(Dataset):
\n def init(self, annotations_file, audio_dir, transformation, target_sample_rate, num_samples, device):
\n self.annotations = pd.read_csv(annotations_file)
\n self.audio_dir = audio_dir
\n self.device = device
\n self.transformation = transformation.to(self.device)
\n self.target_sample_rate = target_sample_rate
\n self.num_samples = num_samples
def __len__(self):\n return len(self.annotations)\n\ndef __getitem__(self, index):\n audio_sample_path = self._get_audio_sample_path(index)\n label = self._get_audio_sample_label(index)\n signal, sr = torchaudio.load(audio_sample_path)\n signal = signal.to(self.device)\n signal = self._resample_if_necessary(signal, sr)\n signal = self._mix_down_if_necessary(signal)\n signal = self._right_pad_if_necessary(signal)\n signal = self._cut_if_necessary(signal)\n signal = self.transformation(signal)\n return signal, label\n\ndef _get_audio_sample_path(self, index):\n file_name = self.annotations.iloc[index, 11]\n path = os.path.join(self.audio_dir, file_name)\n return path\n\ndef _get_audio_sample_label(self, index):\n return self.annotations.iloc[index, 0]\n\ndef _resample_if_necessary(self, signal, sr):\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n signal = resample(signal)\n return signal\n\ndef _mix_down_if_necessary(self, signal):\n if signal.shape[0] > 1:\n signal = torch.mean(signal, dim=0, keepdim=True)\n return signal\n\ndef _cut_if_necessary(self, signal):\n if signal.shape[1] > self.num_samples:\n signal = signal[:, :self.num_samples]\n return signal\n\ndef _right_pad_if_necessary(self, signal):\n length_signal = signal.shape[1]\n if length_signal < self.num_samples:\n num_missing_samples = self.num_samples - length_signal\n last_dim_padding = (0, num_missing_sampels)\n signal = torch.nn.functional.pad(signal, last_dim_padding)\n return signal`\n`class CNNNetwork(nn.Module):
\n def init(self):
\n super().init()
self.conv1 = nn.Sequential(\n nn.Conv2d(\n in_channels=1,\n out_channels=16,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n\n self.conv2 = nn.Sequential(\n nn.Conv2d(\n in_channels=16,\n out_channels=32,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.conv3 = nn.Sequential(\n nn.Conv2d(\n in_channels=32,\n out_channels=64,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.conv4 = nn.Sequential(\n nn.Conv2d(\n in_channels=64,\n out_channels=128,\n kernel_size=3,\n stride=1,\n padding=2\n ),\n nn.ReLU(),\n nn.MaxPool2d(kernel_size=2)\n )\n self.flatten = nn.Flatten()\n self.linear = nn.Linear(in_features= 128 * 5 * 3 , out_features=num_classes)\n self.softmax = nn.Softmax(dim=1)\n\ndef forward(self, input_data):\n x = self.conv1(input_data)\n x = self.conv2(x)\n x = self.conv3(x)\n x = self.conv4(x)\n x = self.flatten(x)\n logits = self.linear(x)\n predictions = self.softmax(logits)\n return predictions`\n`ANNOTATIONS_FILE = '../input/birdclef-2023/train_metadata.csv'
\nAUDIO_DIR = \"../input/birdclef-2023/train_audio/\"
\nSAMPLE_RATE = 32000
\nNUM_SAMPLES = 32000
device = \"cuda\" if torch.cuda.is_available() else \"cpu\"
\nprint(\"Running on\", device)
mel_spectogram = torchaudio.transforms.MelSpectrogram(sample_rate = SAMPLE_RATE, n_fft=2028, hop_length=1024, n_mels=64)
\nbird_ds = BirdDataset(ANNOTATIONS_FILE, AUDIO_DIR, mel_spectogram, SAMPLE_RATE, NUM_SAMPLES, device)`
`from torch.utils.data import DataLoader
\nBATCH_SIZE = 128
\nEPOCHS = 10
\nLEARNING_RATE = 0.001
device = \"cuda\" if torch.cuda.is_available() else \"cpu\"
\ndef create_data_loader(train_data, batch_size):
\n train_dataloader = DataLoader(train_data, batch_size=batch_size)
\n return train_dataloader
def train_single_epoch(model, data_loader, loss_fn, optimiser, device):
\n for input, target in data_loader:
\n input, target = torch.tensor(input).to(device), torch.tensor(target).to(device)
\n # calculate loss
\n prediction = model(input)
\n loss = loss_fn(prediction, target)
# backpropagate error and update weights\n optimiser.zero_grad()\n loss.backward()\n optimiser.step()\n\nprint(f\"loss: {loss.item()}\")\ndef train(model, data_loader, loss_fn, optimiser, device, epochs):
\n for i in range(epochs):
\n print(f\"Epoch {i+1}\")
\n train_single_epoch(model, data_loader, loss_fn, optimiser, device)
\n print(\"---------------------------\")
\n print(\"Finished training\")
train_dataloader = create_data_loader(bird_ds, BATCH_SIZE)
\ncnn = CNNNetwork().to(device)
\nprint(cnn)
loss_fn = nn.CrossEntropyLoss()
\noptimiser = torch.optim.Adam(cnn.parameters(),
\n lr=LEARNING_RATE)
\ntrain(cnn, train_dataloader, loss_fn, optimiser, device, EPOCHS)
\ntorch.save(cnn.state_dict(), \"feedforwardnet.pth\")
\nprint(\"Trained feed forward net saved at feedforwardnet.pth\")`
At first I thought it was extra labels but apparently it's not. I know my code is not optimized at all and the model might create nightmares for advanced people but I'm just trying to learn and get a first prediction to be able to tweak the model after :)
\nI would really appreciate a little boost/help here to what I'm doing wrong :(
\nThank you very much
", "rawMarkdown": "Hi everyone, I'm currently trying my best to develop a CNN to classify the birds sound as a beginner data scientist, but I run into an issue while trying to train my model, here is the entire traceback:\n`---------------------------------------------------------------------------\nValueError Traceback (most recent call last)\n/tmp/ipykernel_27/2595944468.py in