Input 2D matrices and Ouptut scalar

[papipopapu]

Hello! I am new to TensorFlow, so please mind the lack of knowledge. I decided to start a project to improve my skills in which I need to create a model which takes 2D matrixes and outputs scalars. So I just created my array of 2D matrices X_train and the training output y_train.

#1
model_1 = tf.keras.Sequential([                                                            
    tf.keras.layers.Dense(100),
    tf.keras.layers.Dense(10),
    tf.keras.layers.Dense(1, activation = "relu")
])

#2
model_1.compile(optimizer = tf.optimizers.Adam(), loss = tf.keras.losses.mae )
#3
history = model_1.fit(X_train,y_train, epochs = 100, callbacks = [callback])

But I just cant figure out how keras works with layers and I always end up with layers that cant output a scalar. I tried changing the input shape but with no success.

model_1.summary:

As expected, if I do model_1.predict(X_test).shape it gives (1970, 30, 1), when it should have been a vector. For context, X_train is (7970, 30, 5) and y_train is (7970, ). I know I am missing something important. Any help for this newbie is much appreciated.

Thanks!

[ashikshafi08]

Hey @papipopapu

I think you may be looking for something like a Flatten layer, it will condense your outputs at the end.

#1
model_1 = tf.keras.Sequential([                                                            
    tf.keras.layers.Dense(100),
    tf.keras.layers.Dense(10),
   tf.keras.layers.Flatten(), # Add the flatten layer in here
    tf.keras.layers.Dense(1, activation = "relu")
])

#2
model_1.compile(optimizer = tf.optimizers.Adam(), loss = tf.keras.losses.mae )
#3
history = model_1.fit(X_train,y_train, epochs = 100, callbacks = [callback])

[papipopapu]

Thank you, works perfectly!

[mrdbourke]

I was going to answer something very similar to @ashikshafi08!

You could also try the Flatten layer at the start of the network:

#1
model_1 = tf.keras.Sequential([  
    tf.keras.layers.Flatten(), # Add the flatten layer in here                                                          
    tf.keras.layers.Dense(100),
    tf.keras.layers.Dense(10),
    tf.keras.layers.Dense(1, activation = "relu")
])

#2
model_1.compile(optimizer=tf.optimizers.Adam(), loss=tf.keras.losses.mae)

#3
history = model_1.fit(X_train, y_train, epochs=100, callbacks=[callback])

Give it a try and let us know how you go.

[papipopapu]

Yeah, it works like a charm, love u man thanks