[08-seq_classification.ipynb] Creation of the same batch element by Dataset Generator.

[gokulkrishna98]

NYU-DLSP20/res/sequential_tasks.py

Line 224 in 484b2b0

def __getitem__(self, index):

In the following getitem() method, the index variable is not used. So, every batch generated is identical since seed is set by the get_predefined_generator().

Not sure, if this is intended behavior. Essentially, we are training same batch element over (1000/batch_size) iterations and across epochs.

    def __getitem__(self, index):
        batch_x, batch_y = [], []
        for _ in range(self.batch_size):
            x, y = self.generate_pair()
            batch_x.append(x)
            batch_y.append(y)
        return self.encode_x_batch(batch_x), self.encode_y_batch(batch_y)