Trouble using dense_layers argument

[chriskok]

Hi! I've recently been hyperparameter tuning for the OCMAML algorithm on the MNIST dataset and I ran into this error when trying to use the dense_layers argument:

Traceback (most recent call last):
File "main.py", line 640, in
main(args)
File "main.py", line 313, in main
model = MAML(sess, args, seed, n_train_tasks, input_shape)
File "/media/nvidia/DATA1/MUHB/Few-Shot-One-Class-Classification-via-Meta-Learning/MAMLs_Reptiles/MNIST/metalearning_algorithms/maml_class.py", line 209, in init
self.updated_bn_model = self.assign_stats(self.X_finetune)
File "/media/nvidia/DATA1/MUHB/Few-Shot-One-Class-Classification-via-Meta-Learning/MAMLs_Reptiles/MNIST/metalearning_algorithms/maml_class.py", line 862, in assign_stats
assign_op_3 = self.layers[i].variables[-1].assign(var)
File "/home/nvidia/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/resource_variable_ops.py", line 915, in assign
self._shape.assert_is_compatible_with(value_tensor.shape)
File "/home/nvidia/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 1023, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (256,) and (512,) are incompatible

To reproduce: use the given OCMAML config file but with the dense_layers argument set to "256 128 64 64" (as suggested in MNIST/metalearning_algorithms/main.py). Any suggestions on how to fix the issue and add dense layers?

Really appreciate it!

[17011813]

Good Morning, chriskok

How did you solve this problem?

[AhmedFrikha]

Hi, sorry for the late answer. Some changes have to be made to the current code does not allow the usage of fully connected networks. I plan to add this feature soon.

[17011813]

@AhmedFrikha Hello, sir.

When are you gonna upload this feature?
I can't wait to check more operations and explanations about past issue #3

Also, How can I run your OC-MetaSGD or OC-MAML code with STS-Sawtooth dataset or CNC milling machine dataset?

Thank you again :)
Have a great day!