intel-nrc-ecosystem / models Goto Github PK

Hi,

I'm trying to use the NxConv1D. However, I encountered these problems:

1. in dnn_layers.py, class NxConv1D, the call function's line 1410 doesn't have the attribute '_recreate_conv_op';
2. if I comment that line out, then the error becomes: AttributeError: 'NxConv1D' object has no attribute '_padding_op'.

So what are these 2 attributes in NxConv1D class?

Thanks!

Hello,
Thanks for sharing the DNN implementation on the Loihi chip!
When I was trying to reproduce the b_image_classification_cifar.ipynb, I encountered this error:
AttributeError: 'SNN' object has no attribute '_is_aedat_input'.
Any ideas on how to fix this?

Thanks!
`---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
in
7 )
8
----> 9 convert_model(config_file_path)

in convert_model(config_file_path)
4 from snntoolbox.bin.run import main
5
----> 6 main(config_file_path)

~/nxsdk095/lib/python3.5/site-packages/snntoolbox/bin/run.py in main(filepath)
29 if filepath is not None:
30 config = update_setup(filepath)
---> 31 run_pipeline(config)
32 return
33

~/nxsdk095/lib/python3.5/site-packages/snntoolbox/bin/utils.py in run_pipeline(config, queue)
125 config.get('paths', 'filename_parsed_model')))
126
--> 127 spiking_model.build(parsed_model, **testset)
128
129 # Export network in a format specific to the simulator with which it

~/nxsdk095/lib/python3.5/site-packages/snntoolbox/simulation/utils.py in build(self, parsed_model, **kwargs)
434
435 # Iterate over layers to create spiking neurons and connections.
--> 436 self.setup_layers(batch_shape)
437
438 print("Compiling spiking model...\n")

~/nxsdk095/lib/python3.5/site-packages/snntoolbox/simulation/utils.py in setup_layers(self, batch_shape)
765 """Iterates over all layers to instantiate them in the simulator"""
766
--> 767 self.add_input_layer(batch_shape)
768 for layer in self.parsed_model.layers[1:]:
769 print("Building layer: {}".format(layer.name))

~/0.9.5/nxsdk-apps-0.9.5/inrc_dnn/nxsdk_modules_ncl/snntoolbox/nx_backend.py in add_input_layer(self, input_shape)
255 if self._poisson_input:
256 raise NotImplementedError
--> 257 elif self._is_aedat_input:
258 input_mode = nxtf.InputModes.AEDAT
259 else:`

Hi,
Inside the class SigmaDeltaNeuronsInstErrExec, the use of "self.process.act_fn(y) - self.process.act_fn(vars.y_ref)" to compute delta_a causes errors when the activation function is LeakyReLU. This happens because, delta_a is used to update y_ref. Consider a time t when the value of y becomes negative and the slope of leakyrelu for negative values be 0.1, the value of delta_a correctly corresponds to a value (y(t)0.1 - y_ref(t)) and it update y_ref(t) to reflect the value of 0.1y. But in the next time step when y_ref is passed to the activation function, the output becomes 0.01*y(t) and the absolute value of delta_a is more. Instead of applying the activation function over y_ref, a direct subtraction is sufficient and it works for ReLU and other non-linear activation functions also.
I have added the section of code which worked for me.

delta_a = tf.cond(
tf.logical_and(
tf.greater_equal(self.time, self.process.first_valid_ts),
tf.equal(tf.math.mod(self.time
- self.process.first_valid_ts + 1,
self.process.out_decim_interval), 0)),
true_fn=lambda: self.process.act_fn(y)
- vars.y_ref,
false_fn=lambda: tf.zeros_like(vars.delta_a))

Hi,

I am trying to tune an ANN->SNN with a KB device. The example snn_toolbox/examples/mnist_keras_loihi.py tried to run a SLURM session, even though I removed lines 26 os.environ['SLURM'] = '1' and 27 os.environ['PYTHONUNBUFFERED'] = '1'.

Further inspection revealed that models/nxsdk_modules_ncl/snntoolbox/nx_backend.py also sets the SLURM environment in line 71 os.environ['SLURM'] = '1'.

Removing the line lead to successful execution. I thought it might be desired behavior that the environment will be set at top level of the program; maybe you want to remove that line in the repo?