Error after generating ant 1

[isaacgerg]

I get the following error everytime after generating ant 1

builtins.ValueError: Tensor("Adam/iterations:0", shape=(), dtype=resource) must be from the same graph as Tensor("training/Adam/Const:0", shape=(), dtype=int64).

Prior output is:

=======
Ant: 0x5f7a92e8
Loss: 0.531042
Accuracy: 0.750683
Path: InputNode(shape:(256, 256, 1)) -> Conv2DNode(kernel_size:1, filter_count:64, activation:ReLU) -> FlattenNode() -> OutputNode(output_size:1, activation:Sigmoid)
Hash: 2510cfe0ba8648855dc73f4c2cb8e7ff75878eeee906c211c51f470bc6ff0547

---------------------------Current search depth is 1----------------------------
--------------------------------GENERATING ANT 1--------------------------------
Train on 5270 samples, validate on 586 samples

[isaacgerg]

I was able to fix this issue (so far) by having free_gpu() simply return without doing K.clear_session().

[Pattio]

Hmm that's interesting I just ran the test with newest version in Google Colab and it works without any errors. Could you please provide following details:

1. Operating system
2. TensorFlow version
3. Training sample code

[isaacgerg]

1. Windows 7 x64
2. 1.13.1

dataset = Dataset(training_examples=x_train, training_labels=y_train, testing_examples=x_train, testing_labels=y_train)
backend = TFKerasBackend(dataset=dataset, optimizer=tf.keras.optimizers.Adam(1e-4))
deepswarm = DeepSwarm(backend=backend)
topology = deepswarm.find_topology()
trained_topology = deepswarm.train_topology(topology, 50)

the yaml

DeepSwarm:
    save_folder:
    metrics: accuracy
    max_depth: 15
    reuse_patience: 1

    aco:
        pheromone:
            start: 0.1
            decay: 0.1
            evaporation: 0.1
            verbose: False
        greediness: 0.5
        ant_count: 16

    backend:
        epochs: 15
        batch_size: 16
        patience: 5
        loss: binary_crossentropy
        verbose: True

    spatial_nodes: [InputNode, Conv2DNode, DropoutSpatialNode, BatchNormalizationNode, Pool2DNode]
    flat_nodes: [FlattenNode, DenseNode, DropoutFlatNode, BatchNormalizationFlatNode]

Nodes:

    InputNode:
        type: Input
        attributes: 
            shape: [!!python/tuple [256, 256, 1]]
        transitions:
            Conv2DNode: 1.0

    Conv2DNode:
        type: Conv2D
        attributes:
            filter_count: [32, 64, 128]
            kernel_size: [1, 3, 5]
            activation: [ReLU]
        transitions:
            Conv2DNode: 0.8
            Pool2DNode: 1.2
            FlattenNode: 1.0
            DropoutSpatialNode: 1.1
            BatchNormalizationNode: 1.2
    
    DropoutSpatialNode:
        type: Dropout
        attributes:
            rate: [0.1, 0.3]
        transitions:
            Conv2DNode: 1.1
            Pool2DNode: 1.0
            FlattenNode: 1.0
            BatchNormalizationNode: 1.1

    BatchNormalizationNode:
        type: BatchNormalization
        attributes: {}
        transitions:
            Conv2DNode: 1.1
            Pool2DNode: 1.1
            DropoutSpatialNode: 1.0
            FlattenNode: 1.0

    Pool2DNode:
        type: Pool2D
        attributes:
            pool_type: [max, average]
            pool_size: [2]
            stride: [2, 3]
        transitions:
            Conv2DNode: 1.1
            FlattenNode: 1.0
            BatchNormalizationNode: 1.1

    FlattenNode:
        type: Flatten
        attributes: {}
        transitions:
            DenseNode: 1.0
            OutputNode: 0.8
            BatchNormalizationFlatNode: 0.9

    DenseNode:
        type: Dense
        attributes:
            output_size: [64, 128]
            activation: [ReLU, Sigmoid]
        transitions:
            DenseNode: 0.8
            DropoutFlatNode: 1.2
            BatchNormalizationFlatNode: 1.2
            OutputNode: 1.0

    DropoutFlatNode:
        type: Dropout
        attributes:
            rate: [0.1, 0.3]
        transitions:
            DenseNode: 1.0
            BatchNormalizationFlatNode: 1.0
            OutputNode: 0.9

    BatchNormalizationFlatNode:
        type: BatchNormalization
        attributes: {}
        transitions:
            DenseNode: 1.1
            DropoutFlatNode: 1.1
            OutputNode: 0.9

    OutputNode:
        type: Output
        attributes:
            output_size: [1]
            activation: [Sigmoid]
        transitions: {}

[Pattio]

Thank you, I located the issue and will update you once it has been fixed.

[isaacgerg]

Sure. Also, if you update all your open(WindowPath) calls to open(str(WindowPath)) you should be python 3.5 compatible. open() and Pathlib are only seamless in 3.6.

[Pattio]

Issue is now fixed in version 0.0.7. Regarding the Python version, thank you for the suggestion, but when developing the library decision was made to aim for Python 3.6, without any backward compatibility (as this allows easier development). At the moment I don't really see any problem with using 3.6 as the default version.