Introduces the key components of the DimeNet++ model to the HydraGNN library by modifying the PyG implementation.
The core addition is the DIMEStack which is inherits from Base to provide convolutional layers on request using the _get_conv() method.
Notice: Because the PyG implementations use the glorot initialization, the construction of the hidden dimension must be larger than 1 to avoid division by zero. Only the input_dim and output_dim are provided to _get_conv(), without constraint on the dimension of the convolution. Therefore, DIMEStack will use the input_dim if larger than 1 or the output_dim otherwise. If neither is larger than one, then the assertion will fail.
The DIMEStack also computes the necessary convolutional arguments, being
rbf: (tensor) the radial basis function output.
sbf: (tensor) the spherical basis function output.
i: The message passing target.
j: The message passing source.
idx_kj: A portion of the bond triple.
idx_ji: The second component of the bond triple.
Notice: DimeNet uses the triplet() method to compute the bond angles between triplets of nodes. This information is computed on every call but there is room for computational speed ups by computing once and then updating iteratively.
The stack modifies the Embedding , the baseline atomic feature embedding is removed and embedding is handled by HydraGNN.
Eight hyperparameters are introduced.
basis_emb_size/int_emb_size/out_emb_size determines the dimensions of the embedding mechanism.
num_before_skip/num_after_skip control layers before and after the attention mechanism in the OutputBlock.
num_radial/envelope_exponent determine the radial basis functions used.
num_spherical determines the number of spherical basis functions used
Introduces the key components of the DimeNet++ model to the HydraGNN library by modifying the PyG implementation.
The core addition is the DIMEStack which is inherits from
Baseto provide convolutional layers on request using the_get_conv()method.Notice: Because the PyG implementations use the glorot initialization, the construction of the hidden dimension must be larger than 1 to avoid division by zero. Only the
input_dimandoutput_dimare provided to_get_conv(), without constraint on the dimension of the convolution. Therefore, DIMEStack will use theinput_dimif larger than 1 or theoutput_dimotherwise. If neither is larger than one, then the assertion will fail.The DIMEStack also computes the necessary convolutional arguments, being
rbf: (tensor) the radial basis function output.sbf: (tensor) the spherical basis function output.i: The message passing target.j: The message passing source.idx_kj: A portion of the bond triple.idx_ji: The second component of the bond triple.Notice: DimeNet uses the
triplet()method to compute the bond angles between triplets of nodes. This information is computed on every call but there is room for computational speed ups by computing once and then updating iteratively.The stack modifies the
Embedding, the baseline atomic feature embedding is removed and embedding is handled by HydraGNN.Eight hyperparameters are introduced.
basis_emb_size/int_emb_size/out_emb_sizedetermines the dimensions of the embedding mechanism.num_before_skip/num_after_skipcontrol layers before and after the attention mechanism in theOutputBlock.num_radial/envelope_exponentdetermine the radial basis functions used.num_sphericaldetermines the number of spherical basis functions usedOther updates include documentation and testing.