Descriptor inference according to an trained DP

[wangcmphy]

Hello,
I have trained a DP with se_e2_a descriptor. Now, I would like to use this DP to infer the descriptor tensor for each atomic local environment. How can I do? Thanks!

[anyangml]

You can extract intermediate output by using a hook.

for example

intermediate_output = None

# Define a hook for the specific layer
def hook_fn(module, input, output):
    global intermediate_output
    intermediate_output = output

# Register the hook on the specific layer using its path
# You can modify the path below to match the exact path of the layer you're interested in, in your case, this would be the descriptor
layer_path = "atomic_model.fitting_net.filter_layers.networks.0.layers.2.activate"
hook = mm.get_submodule(layer_path).register_forward_hook(hook_fn)

# Perform a forward pass through the model
output = mm(input_coord, input_atype)

# Now 'intermediate_output' contains the output of the specified layer
print(intermediate_output)

# Don't forget to remove the hook after you're done to avoid side effects
hook.remove()

[njzjz]

Please use DeepEval.eval_descriptor.