possible bug in keras.utils.Sequence

[pure-rgb]

Code

from tensorflow import keras
import numpy as np

X_test_data = [np.random.rand(20, 20, 1).astype(np.float32) for _ in range(5)]
y_test_data = [np.random.randint(0, 2, (20, 20,  1)).astype(np.float32) for _ in range(5)]

class SimpleDataLoader(keras.utils.Sequence):
    def __init__(
        self,
        x_data,
        y_data,
        batch_size=1,
        shuffle=True,
    ):
        self.x_data = x_data
        self.y_data = y_data
        self.batch_size = batch_size
        self.shuffle = shuffle
        self.on_epoch_end()

    def __len__(self):
        return int(np.floor(len(self.x_data) / self.batch_size))

    def __getitem__(self, index):
        print(f"__getitem__ called with index: {index})")
        indices = self.indices[index * self.batch_size:(index + 1) * self.batch_size]
        x_batch = [self.x_data[k] for k in indices]
        y_batch = [self.y_data[k] for k in indices]

        if not x_batch:
            print(f"Warning: x_batch is empty for index {index})")
        if not y_batch:
            print(f"Warning: y_batch is empty for index {index})")

        X = np.stack(x_batch, axis=0)
        y = np.stack(y_batch, axis=0)

        return X, y

    def on_epoch_end(self):
        self.indices = np.arange(len(self.x_data))
        if self.shuffle:
            np.random.shuffle(self.indices)

batch_size = 1

vl = SimpleDataLoader(
    x_data=X_test_data,
    y_data=y_test_data,
    batch_size=1,
    shuffle=False,
)

print(len(vl))
for x, y in vl:
    print(f"Shape: x={x.shape}, y={y.shape}")

The length of the input is simply 5, if for validation, it should stop after iterating over the dataset but it keeps trying more and ended up error.

5
__getitem__ called with index: 0)
Shape: x=(1, 20, 20, 1), y=(1, 20, 20, 1)
__getitem__ called with index: 1)
Shape: x=(1, 20, 20, 1), y=(1, 20, 20, 1)
__getitem__ called with index: 2)
Shape: x=(1, 20, 20, 1), y=(1, 20, 20, 1)
__getitem__ called with index: 3)
Shape: x=(1, 20, 20, 1), y=(1, 20, 20, 1)
__getitem__ called with index: 4)
Shape: x=(1, 20, 20, 1), y=(1, 20, 20, 1)
__getitem__ called with index: 5)
Warning: x_batch is empty for index 5)
Warning: y_batch is empty for index 5)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
/tmp/ipykernel_31/1498463694.py in <cell line: 0>()
     54 
     55 print(len(vl))
---> 56 for x, y in vl:
     57     print(f"Shape: x={x.shape}, y={y.shape}")

/tmp/ipykernel_31/1498463694.py in __getitem__(self, index)
     34             print(f"Warning: y_batch is empty for index {index})")
     35 
---> 36         X = np.stack(x_batch, axis=0)
     37         y = np.stack(y_batch, axis=0)
     38 

/usr/local/lib/python3.11/dist-packages/numpy/core/shape_base.py in stack(arrays, axis, out, dtype, casting)
    443     arrays = [asanyarray(arr) for arr in arrays]
    444     if not arrays:
--> 445         raise ValueError('need at least one array to stack')
    446 
    447     shapes = {arr.shape for arr in arrays}

ValueError: need at least one array to stack

[pure-rgb]

Raised before.

#19994
#20001

Why the hell hasn’t this been fixed yet?