Random sample from PrefetchDataset 07_Food_vision_big

[Natali-bali]

Hello again, we used .take(1) method to check one (first) element from PrefetchDataset (format of downloaded data from tfds).
If i use take(10) it would return TakeDataset with 10 first elements from PrefetchDataset.
My question is how i can get 10 random elements from PrefetchDataset?
Thanks a lot)

[gauravreddy08]

Hey @Natali-bali,

I was also trying to figure out this. I used this below code for my project :

import matplotlib.pyplot as plt

plt.figure(figsize=(10, 10))

for i in range(9):
  for images, labels in test_data.take(1):
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(images.numpy().astype("uint8"))
    x=labels.numpy()

    pred_prod = model.predict(tf.expand_dims(images, axis=0))
    pred = class_names[np.round(int(pred_prod[0]))]
    
    if pred==class_names[int(x)]: plt.title(f"{pred}", color='white', bbox=dict(boxstyle="round", color='green'))
    else: plt.title(f"{pred}", color='white', bbox=dict(boxstyle="round", color='red'))

    plt.axis("off")

Here's a simplified version :

for i in range(10):
  for images, labels in test_data.take(1):
    do_something()

[gauravreddy08]

It's strange that it worked for everyone. Can u share the notebook link ?

[AbdullahMansoor123]

It's strange that it worked for everyone. Can u share the notebook link ?

https://colab.research.google.com/drive/16Aphr8nrvkzxC2uED6XOs7mTkWTl2avP?usp=sharing

check the Question#1c

[Natali-bali]

for images, label in test_data.take(1):
plt.show(image)

[AbdullahMansoor123]

for images, label in test_data.take(1):
plt.show(image)

not this is not the issue

[AbdullahMansoor123]

@gauravreddy08
I was prefetching and batching my data from the start even though I didn't need to as my model was already trained ,the right way was to take test data directly from tfds function and then preprocess data like you usually do for custom images and plot it,

@Natali-bali But the problem still remains how to randomize the images