AttributeError: 'stochastic_ternary' object has no attribute 'shape' --- happened when using Conv1D

[Duchstf]

Hi,

I was trying to use the QConv1D layer and encounter this error. Does anyone have any idea why?

Traceback (most recent call last):
  File "quantize-Conv1D.py", line 179, in <module>
    UseNetwork(args.weight_file, save_model = args.save_model, load_weights=lw)
  File "quantize-Conv1D.py", line 142, in UseNetwork
    model = QConv1D_model(weights_f, load_weights)
  File "quantize-Conv1D.py", line 101, in QConv1D_model
    x = QConv1D(filters=128, kernel_size=3, kernel_quantizer="stochastic_ternary", bias_quantizer="ternary", name="conv1d_1")(x)
  File "/home/duchstf/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 842, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "/home/duchstf/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/autograph/impl/api.py", line 237, in wrapper
    raise e.ag_error_metadata.to_exception(e)
AttributeError: in converted code:
    relative to /home/duchstf/miniconda3/lib/python3.7/site-packages:

    qkeras/qconvolutional.py:127 call  *
        outputs = tf.keras.backend.conv1d(
    tensorflow_core/python/keras/backend.py:4804 conv1d
        kernel_shape = kernel.shape.as_list()

    AttributeError: 'stochastic_ternary' object has no attribute 'shape'

This is how the QConv1D layer was used in my code

x = x_in = Input(IN_SHAPE, name="input")
x = QConv1D(filters=128, kernel_size=3, kernel_quantizer="stochastic_ternary", bias_quantizer="ternary", name="conv1d_1")(x)

Thanks,

Duc.

[nunescoelho]

We are enhancing the quantizers to accept just the name, like in Keras. Meanwhile, try to do "stochastic_ternary()" and "ternary()".

Best regards,

Claudionor

On Mon, Feb 3, 2020 at 9:32 AM Duc Hoang notifications@github.com wrote:

Hi,

I was trying to use the QConv1D layer and encounter this error. Does anyone have any idea why?

Traceback (most recent call last): File "quantize-Conv1D.py", line 179, in UseNetwork(args.weight_file, save_model = args.save_model, load_weights=lw) File "quantize-Conv1D.py", line 142, in UseNetwork model = QConv1D_model(weights_f, load_weights) File "quantize-Conv1D.py", line 101, in QConv1D_model x = QConv1D(filters=128, kernel_size=3, kernel_quantizer="stochastic_ternary", bias_quantizer="ternary", name="conv1d_1")(x) File "/home/duchstf/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 842, in call outputs = call_fn(cast_inputs, *args, **kwargs) File "/home/duchstf/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/autograph/impl/api.py", line 237, in wrapper raise e.ag_error_metadata.to_exception(e) AttributeError: in converted code: relative to /home/duchstf/miniconda3/lib/python3.7/site-packages:

qkeras/qconvolutional.py:127 call  *
    outputs = tf.keras.backend.conv1d(
tensorflow_core/python/keras/backend.py:4804 conv1d
    kernel_shape = kernel.shape.as_list()

AttributeError: 'stochastic_ternary' object has no attribute 'shape'

This is how the QConv1D layer was used in my code

x = x_in = Input(IN_SHAPE, name="input") x = QConv1D(filters=128, kernel_size=3, kernel_quantizer="stochastic_ternary", bias_quantizer="ternary", name="conv1d_1")(x)

Thanks,

Duc.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/google/qkeras/issues/13?email_source=notifications&email_token=AM5ENPH2DKRTGTURDOJR56LRBBIKPA5CNFSM4KPJDQLKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IKVEWKQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5ENPGW54A57AIBJIINNHDRBBIKPANCNFSM4KPJDQLA .

[Duchstf]

Thank you so much! It works now! There is still a minor problem regarding print_qstats. However, everything works fine if I comment it out.

Traceback (most recent call last):
  File "quantize-Conv1D.py", line 179, in <module>
    UseNetwork(args.weight_file, save_model = args.save_model, load_weights=lw)
  File "quantize-Conv1D.py", line 142, in UseNetwork
    model = QConv1D_model(weights_f, load_weights)
  File "quantize-Conv1D.py", line 131, in QConv1D_model
    print_qstats(model)
  File "/home/duchstf/miniconda3/lib/python3.7/site-packages/qkeras/estimate.py", line 493, in print_qstats
    model_ops = extract_model_operations(model)
  File "/home/duchstf/miniconda3/lib/python3.7/site-packages/qkeras/estimate.py", line 415, in extract_model_operations
    kernel_h, kernel_w, _, _ = weight.shape
ValueError: not enough values to unpack (expected 4, got 3)

Thanks,

Duc.

[Duchstf]

Created PR to fix the problem. Thanks!