AI4M C1W1 Weighted_loss "All test passed" data type issue

Course Q&A AI For Medicine AI for Medical Diagnosis
1

Hi ,
I am facing two issues but it’s likely that both issues are related something to do with using tensor and numpy in the same equation?

Issue 1: All my tests pass but during grading it complains about datatype in weighted_loss function while calculating the final loss value.

Issue 2: It also seems like K.mean function doesn’t work, so I have to manually divide it by total samples.
I’ve explained each issue below

Issue 1:

Attempt1:

    loss = -K.mean(loss/len(y_true))
    print(loss.dtype)

    *Your outputs:*
      L(y_pred_1) =  -0.49562032355455976
      L(y_pred_2) =  -0.49562032355455976
      Difference: L(y_pred_1) - L(y_pred_2) =  0.0 
      <dtype: 'float64'>
      Error: Data-type mismatch. Make sure it is a np.float32 value.

Attempt 2:

  loss = tf.cast(-K.mean(loss/len(y_true)),tf.float32)
  print(loss.dtype)

  *Your outputs:*
  L(y_pred_1) =  -0.4956203
  L(y_pred_2) =  -0.4956203
  Difference: L(y_pred_1) - L(y_pred_2) =  0.0 
  <dtype: 'float32'>
   All tests passed.

HOWEVER, this is what I get when I submit attempt 2 for grading:

"Unexpected error occurred during function check. We expected function `get_weighted_loss` to return a function, 'weighted_loss', and the 'weighted_loss' function should return a type <class 'float'>. Please check that this function is defined properly. "

Issue 2:
loss = -K.mean(loss) calculates wrong value for loss

  Your outputs:
  L(y_pred_1) =  -1.982481294218239
  L(y_pred_2) =  -1.982481294218239
  Difference: L(y_pred_1) - L(y_pred_2) =  0.0 
  Error: Wrong output. One possible mistake, your epsilon is not equal to 1.

Thanks for any help :slight_smile:

2

I would like to comment on issue 2.

The variable “loss” should be the float type. However, inferring from your code snippet, you passed the “loss” to K.mean(). The arguments of K.mean() are supposed to be tensor so that you can’t use K.mean().

It is possible that each variable type is NOT the one it is expected, so it might be a good idea to check the variable type of each variable.

2 Likes
4

That was it. Thank you!
Tried to solve everything using numpy since I was more familiar with numpy than tensorflow. Did away with numpy completely and grading passed as well. Thanks again :slightly_smiling_face:

1 Like