Chinese Dragon Box

thingiverse

Based on the code snippet provided, it appears that you're attempting to create a Conv2D neural network for image classification. However, there are several issues with your implementation. Firstly, the shape of `x_train` and `y_train` do not match the expected input shape for Conv2D layer (3 dimensional images with depth as 1, but here it seems like all values in arrays represent only the coordinate positions). Also, the output is 28 units because the dense function before this has units = 14 x 14, yet these layers should output unit(s) equal to classes of problem you want solution to - and since no activation (activation="softmax" for a classifcation), softmax or equivalent softmax for multiple classifications can be assumed so in essence we expect shape: number_of_classes. To rectify this: - The dimensions don't align between input x and weights/featuremaps. The images expected for your problem is different (3 channels color etc... but 28 x 28). ```python model = Sequential([ Conv2D(32, (3,3), activation='relu', input_shape=(28, 28, 1)), Flatten(), Dense(64, activation='relu'), Dense(10, activation='softmax') ]) ``` This would mean you need images in `x_train` with dimensions like (n_samples x 28 x 28 x 3). To create an instance of the network as defined here you will use the function Keras API provides called 'build': ```python def build(input_shape): model = Sequential([ Conv2D(32, (3,3), activation='relu', input_shape=input_shape), Flatten(), Dense(64, activation='relu'), Dense(n_classes, activation='softmax') ]) return model model = build((28, 28, 1)) ```