Quick question - urgent help!

[JoshK68]

Hi friends,

Taking this exceptional class right now and ran into an error. Was running this on 2 Food classes.
Here's the code and the error.
Can you please advise on what I did wrong and how I can quickly fix it.
Thank you very much!
Josh

def create_model(model_url, num_classes=2):
'''Takes a TensorFlow Hub URL and creates a Keras Sequential model with it.

Args:
model_url (str): A TF Hub feature extraction URL.
num_classes(int): Number of output neurons in the outlayer,
should be equal to number of target classes, default 10.

Returns:
An uncopiled Keras Sequential model with model_url as feature extractor
layer and Dense output layer with num_classes output neurons
'''

#Download the pretrained model and save it as a Keras layer
#On TF hub all the models have example usage
#freeze the already learned patterns - these model have laready ben trained on ImageNet
#give it a name so that we know what it is
feature_extractor_layer = hub.KerasLayer(model_url,
trainable=False,
name='feature_extraction_layer',
input_shape = IMAGE_SHAPE +(3,)) ### Image_shape + color channels

#Create our own model
model = tf.keras.Sequential([
feature_extractor_layer,
layers.Dense(num_classes, activation='sigmoid', name='output_layer') #changed to sigmoid b/c working with binary data

])

return model

#Create EfficientNet model
efficientnet_model = create_model(efficient_net, 2)

#Compile our efficientnet model
efficientnet_model.compile(loss='binary_crossentropy',
optimizer=tf.keras.optimizers.Adam(),
metrics=['accuracy'])

Let's fit our efficientnet model to the data

efficientnet_history = efficientnet_model.fit(train_data,
epochs=5,
steps_per_epoch=len(train_data),
validation_data=test_data,
validation_steps=len(test_data))

ValueError: logits and labels must have the same shape ((None, 2) vs (None, 1))

#where is this shape error coming from and how do I fix it? Been working on it for very long!

Thank you!

[gauravreddy08]

Hey @JoshKonoff,

I guess the error is the num_classes input you are providing to the model which is 2.

Even though you working with a two-class classification scenario, it comes under a binary classification which means that the end neuron should be one.

Solution : So use 1 instead of 2, this should fix the error.

Let me know if this doesn't fix the error !!

[JoshK68]

Hi Gaurxvreddy,

Huge thank you for this!

I changed: def create_model(model_url, num_classes=1):
so that the number of classes would = 1

It still gives me the same error: ValueError: logits and labels must have the same shape ((None, 2) vs (None, 1))

Any advice on what's going on?

Thank you again!

Josh

[gauravreddy08]

Can you share your notebook link ?

[gauravreddy08]

efficientnet_model = create_model(efficient_net, 2)

BTW, change it here too

[JoshK68]

@gaurxvreddy, huge thank you, you got it!

As soon as I changed it here: efficientnet_model = create_model(efficient_net, 2)
it worked.

Thank you very much!
Josh

[gauravreddy08]

Glad that I could help !!

Usually in a Binary Classification problem output neuron is 1. Let's take an example of the Pizza/Steak problem.

• If the output neuron value is between 0.0 and 0.5 its Pizza
• If the output neuron value is between 0.5 and 1.0 its Steak

While on the other hand, Multi-class Classification uses several output neurons, where the no. output neurons = no. of classes

You could also use 2 output neurons in a binary classification problem. but you need to change the model's loss function from binary_crossentropy to categorical_crossentropy

[JoshK68]

Hi Gaurxvreddy,

Quick question...
I created a super simple model and am trying to find the best learning rate.
This is what the error says:

Plot the learning rate versus the loss

lrs = 1e-4 * (10 ** (np.arange(100)/20))
plt.figure(figsize=(10, 7))
plt.semilogx(lrs, history_3.history["loss"]) # we want the x-axis (learning rate) to be log scale
plt.xlabel("Learning Rate")
plt.ylabel("Loss")
plt.title("Learning rate vs. loss");

ValueError: x and y must have same first dimension, but have shapes (100,) and (5,)

When I tried to get the ndims of x and y then it crashed.
What's causing these errors and what advice do you have?

https://colab.research.google.com/drive/1ZeYLIXVJliOL5QAS2VuDbOPZ3khmb_Rt?usp=sharing

Thank you very much!

Josh

[gauravreddy08]

Hey @JoshK68,

I guess you're referring to this code snippet from Daniel's notebook :

If then, here Daniel had trained his model for 100 epochs where in your case you had trained a different model for only 5 epochs.

So you need to have only 5 X-axis values, for that: replace lrs = 1e-4 * (10 ** (np.arange(100)/20)) with lrs = 1e-4 * (10 ** (np.arange(5)/20))

---

When I tried to get the ndims of x and y then it crashed.
What's causing these errors and what advice do you have?

Here the x and y, your referring to is completely different, I guess what you're referring to is the X and y used in training a model. But the x and y in error refer to the values you passed in plt.semilogx(). Where x denotes lrs and y denotes history_3.history["loss"].

[JoshK68]

Hi @gaurxvreddy ,

Yes, you are totally right!
Stupid mistake on my part.
Makes total sense now.

Thank you very much! :)
Josh