[4.2.5] Dive into Deep Learning : exercise answers

4.2. The Image Classification Dataset — Dive into Deep Learning 1.0.3 documentation

 

[1]

Usually batch_size's are formed as $2^n$ from, we can inspect the run time of loading the dataset when batch_size=$2^n$.

ls=[(2*i)**2 for i in range(1, 9)]

t_info = []
for size in ls:
    data = FashionMNIST(resize=(32, 32), batch_size=size)
    tic = time.time()
    for X, y in data.train_dataloader():
        continue
    t_info.append(time.time() - tic)
class ProgressBoard(d2l.HyperParameters):  #@save
    """The board that plots data points in animation."""
    def __init__(self, xlabel=None, ylabel=None, xlim=None,
                ylim=None, xscale='linear', yscale='linear',
                ls=['-', '--', '-.', ':'], colors=['C0', 'C1', 'C2', 'C3'],
                fig=None, axes=None, figsize=(3.5, 2.5), display=True):
        self.save_hyperparameters()

    def draw(self, x, y, label, every_n=1):
        raise NotImplemented
import numpy as np
board = d2l.ProgressBoard('size(exponent 2)')
for x in np.arange(1, 9):
    board.draw(x, t_info[x-1], 'time', every_n=1)

 

result >>

 

We can see that the loading time decreases as the batch_size gets bigger.

---

[2]

Maybe something to do with the hardware.

---

[3]

Datasets — Torchvision 0.17 documentation