No data provided for "input_1"

[GKalliatakis]

I've tried to replicate your build_classification_model using the following code

def build_default_classification_model():
    """Instantiates the initial classification model
    """

    input_shape = [IMG_DIM, IMG_DIM, NUM_BANDS]


    img_input = Input(shape=input_shape)

    # Block 1
    x = Conv2D(filters=32, kernel_size=3, strides=(1, 1), padding='same',
               kernel_regularizer=l2(0.0002),
               activation='relu', name='droughtwatch_block1_conv1')(img_input)

    x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), name="droughtwatch_block1_pool", padding='valid')(x)


    # Block 2
    x = Conv2D(filters=64, kernel_size=3, strides=(1, 1), padding='same',
               kernel_regularizer=l2(0.0002),
               activation='relu', name='droughtwatch_block2_conv1')(x)

    x = Conv2D(filters=64, kernel_size=3, strides=(1, 1), padding='same',
               kernel_regularizer=l2(0.0002),
               activation='relu', name='droughtwatch_block2_conv2')(x)

    x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), name="droughtwatch_block2_pool", padding='valid')(x)

    x = Dropout(0.2, name='droughtwatch_drop_block2')(x)


    # Block 3
    x = Conv2D(filters=64, kernel_size=3, strides=(1, 1), padding='same',
               kernel_regularizer=l2(0.0002),
               activation='relu', name='droughtwatch_block3_conv1')(x)

    x = Conv2D(filters=64, kernel_size=3, strides=(1, 1), padding='same',
               kernel_regularizer=l2(0.0002),
               activation='relu', name='droughtwatch_block3_conv2')(x)

    x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), name="droughtwatch_block3_pool", padding='valid')(x)
    x = Dropout(0.2, name='droughtwatch_drop_block3')(x)
    x = Flatten(name='droughtwatch_flatten')(x)

    # Classification block
    x = Dense(128, activation='relu', name='droughtwatch_fc1')(x)
    x = Dense(50, activation='relu', name='droughtwatch_fc2')(x)

    x = Dense(NUM_CLASSES, activation='softmax', name='droughtwatch_predictions')(x)


    # Create model
    model = Model(img_input, x)



    # set up optimizer
    model.compile(loss=tf.keras.losses.categorical_crossentropy,
                  optimizer=optimizers.Adam(),
                  metrics=['accuracy'])

    return model

And then fit the data using the following:

def train_convnet(data_path,
                  epochs,
                  batch_size=64,
                  num_train=86317,
                  num_val=10778):

    # load training data in TFRecord format
    train_tfrecords, val_tfrecords = load_data(data_path)

    # load images and labels from TFRecords
    train_images, train_labels = parse_tfrecords(train_tfrecords, batch_size, num_train)
    val_images, val_labels = parse_tfrecords(val_tfrecords, batch_size, num_val)

    model = build_default_classification_model()

    # number of steps per epoch is the total data size divided by the batch size
    train_steps_per_epoch = int(math.floor(float(NUM_TRAIN) / float(batch_size)))
    val_steps_per_epoch = int(math.floor(float(NUM_VAL) / float(batch_size)))

    model.fit(train_images, train_labels,
              steps_per_epoch=train_steps_per_epoch,
              epochs=epochs,
              class_weight=class_weights_matrix(),
              validation_data=(val_images, val_labels),
              validation_steps=val_steps_per_epoch)

But I end up with the following error:

in standardize_input_data
    'for each key in: ' + str(names))
ValueError: No data provided for "input_1". Need data for each key in: ['input_1']

Any idea what's wrong with this?

[GKalliatakis]

Update on my question above: training works ok when the sequential model is used instead, but still no with the functional model I shared above.