How to understand tensorflow predictions?

Question

I learning tensorflow from beginning from youtube: https://www.youtube.com/watch?v=Cq_P8kJgjvI&t=1808s

And the last code about predictions like this:

first:

print(df_test_new.iloc[0])
print('prediction')
print(predictions[0])

and the result:

age                              25
workclass                   Private
fnlwgt                       226802
education                      11th
education_num                     7
marital               Never-married
occupation        Machine-op-inspct
relationship              Own-child
race                          Black
sex                            Male
capital_gain                      0
capital_loss                      0
hours_week                       40
native_country        United-States
label                             0
new                             625
Name: 0, dtype: object
prediction
{
  'logits': array([-954.60187], dtype=float32), 
  'logistic': array([0.], dtype=float32), 
  'probabilities': array([1., 0.], dtype=float32), 
  'class_ids': array([0]), 
  'classes': array([b'0'], dtype=object), 
  'all_class_ids': array([0, 1], dtype=int32), 
  'all_classes': array([b'0', b'1'], dtype=object)
}

this is in array index 3

print(df_test_new.iloc[3])
print('prediction:')
print(predictions[3])

and this is the print result:

age                               44
workclass                    Private
fnlwgt                        160323
education               Some-college
education_num                     10
marital           Married-civ-spouse
occupation         Machine-op-inspct
relationship                 Husband
race                           Black
sex                             Male
capital_gain                    7688
capital_loss                       0
hours_week                        40
native_country         United-States
label                              1
new                             1936
Name: 3, dtype: object

prediction:

{
   'logits': array([1222.3406], dtype=float32), 
   'logistic': array([1.], dtype=float32), 
   'probabilities': array([0., 1.], dtype=float32), 
   'class_ids': array([1]), 
   'classes': array([b'1'], dtype=object), 
   'all_class_ids': array([0, 1], dtype=int32), 
   'all_classes': array([b'0', b'1'], dtype=object)
}

I still don't understand prediction meaning in tensorflow please help me to understand it.

Answer 1

From your examples I see the following:

You have data about people and every person has a label, either 1 or 0. This is indicated in your ground truth data under label. What those labels are representing should be explained where ever you got the data from, maybe it's data by a bank if this person should get a loan or not, where the label 1 indicates YES and label 0 indicates NO.

Your prediction overview lists the following information: all_classes list what outputs your neural net can actually deliver, in your case one of two classes (0 or 1). The concrete prediction of what your network is putting out is listed under classes. In your first case, this is b'0' and in your second example this is b'1'. Those predictions indicate 0 and 1 respectively, which means that your network put out the correct prediction for your two examples since the ground truth listed in the data is 0 for the first person and 1 for the second person.