Model.predict() output

So I have this model which is pretrained by Google and also trained on a tfds dataset:

import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow_datasets as tfds

keras = tf.keras

# split the data manually into 80% training, 10% testing, 10% validation
(raw_train, raw_validation, raw_test), metadata = tfds.load(
    split=['train[:80%]', 'train[80%:90%]', 'train[90%:]'],

Since the sizes of our images are all different, we need to convert them all to the same size. 
We can create a function that will do that for us below.

IMG_SIZE = 160  # All images will be resized to 160x160

def format_example(img, lab):
    returns an image that is reshaped to IMG_SIZE
    img = tf.cast(img, tf.float32)
    img = (img / 127.5) - 1
    img = tf.image.resize(img, (IMG_SIZE, IMG_SIZE))
    img = tf.expand_dims(img, axis=0)
    return img, lab

# Now we can apply this function to all our images using .map().
# .map() takes every single example in raw_train, raw_validation, etc and applies the function to it
train =
validation =
test =


# Finally we will shuffle and batch the images
train_batches = train.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
validation_batches = validation.batch(BATCH_SIZE)
test_batches = test.batch(BATCH_SIZE)


# Create the base model from the pre-trained model MobileNet V2
base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
                                               include_top=False,  # we don't want to load the top layer

base_model.trainable = False

Adding OUR Classifier

global_average_layer = tf.keras.layers.GlobalAveragePooling2D()

# Finally, we will add the prediction layer that will be a single dense neuron. We can do this because we
# only have two classes to predict for.
prediction_layer = keras.layers.Dense(1, activation='sigmoid')

# Now we will combine these layers together in a model.
model = tf.keras.Sequential([

Now we will train and compile the model. We will use a very small learning rate to ensure that the model 
does not have any major changes made to it.
base_learning_rate = 0.0001

# Now we can train it on our images
initial_epochs = 10
history =,

 acc = history.history['accuracy']

# you can save a model and load a model with this syntax so you don't have to train it again every time"dogs_vs_cats.h5")  # we can save the model and reload it at anytime in the future
new_model = tf.keras.models.load_model('dogs_vs_cats.h5')
predictions = new_model.predict(test)

get_label_name = metadata.features['label'].int2str  # creates a function object that we can use to get labels

i = 0
for image, label in raw_train.take(6):
    i += 1

Anyway, I am getting this output and I don’t know how to use it or understand it:


In the code, I print the label name and the prediction beneath it, but i dont understand what the number inside the array means and how to use it. I would apreciate it if someone could explain.
Thamk you!

Hello @solgeller3
This is the probability of every single output described in the model.
In many classification models you have a threshold. The common threshold is 0.5, if the result is above of this, then is more likely to be “true”. On the contrary if the result is below is more likely to be false.
The more easy to understand example of this mechanism is the logistic function.


In the specific context of your model, you are making a classification model with images,
and after 10 iterations and at a base learning rate of 0.0001 only the third element,

has a strong probability of being classified correctly.
To optimize the performance of your model:

  1. You can adjust the learning rate. For correctly doing that you must be aware of terms like Gradient descent, Loss Function and Hyper parameter(The learning rate is a Hyperparameter).
  2. You can change the number of iterations, since a more number of iterations could make a prediction more accurate.
    However there is always a risk of over-fitting , this is called bias-variance tradeoff.

But…, In simple words, these are the probabilities that your model (at a given learning rate, iterations, and parameters) will classify correctly every one of your images.
If you want more information about classification threshold:

  1. The basic concept of threshold
    Classification: Thresholding  |  Machine Learning Crash Course
  2. Additional exercises on keras for having fun
    Classification metrics based on True/False positives & negatives
    Hope this helps.
1 Like

I forgot it, even in this model you are using the sigmoid function to make a prediction.
The logistic function is part of the family of sigmoid function.

So a more geometrical explanation would be, the threshold is 0.5 and you can locate every individual output value within the plane:

Hope this helps.

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.