Food101 CNN Tensorflow Playground Model - Flatten layer

[tiagopatriciosantos]

Hello @mrdbourke ,

The model achieves 50% accuracy:

Create a model to replicate the TensorFlow Playground model

model_2 = tf.keras.Sequential([
tf.keras.layers.Flatten(input_shape=(224, 224, 3)), # dense layers expect a 1-dimensional vector as input
tf.keras.layers.Dense(4, activation='relu'),
tf.keras.layers.Dense(4, activation='relu'),
tf.keras.layers.Dense(1, activation='sigmoid')
])

https://dev.mrdbourke.com/tensorflow-deep-learning/03_convolutional_neural_networks_in_tensorflow/#an-end-to-end-example

I've changed the Flatten layer to before the output layer and achieved more than 80% accuracy and 75% in validation:
model_2 = tf.keras.layers.Sequential([
tf.keras.layers.Dense(4,activation="relu", input_shape=(224, 224, 3)),
tf.keras.layers.Dense(4, activation="relu"),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(1, activation="sigmoid")
])

Is my approach incorrect?

Another question, should we look more to improve accuracy, validation accuracy or both?

By the way, great course and resources that you have made for this. Thank you.

[mrdbourke]

Woah! That looks like a big improvement.

If you've validated the model and it works, then your method is correct.

What you'll find when building neural networks is there are very many different ways to do something.

And often tweaking something, even a single layer, can lead to better results.

Fantastic experimenting.

[tiagopatriciosantos]

Wow, thanks!
Seems i'm getting there 😁