models/common.py

import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.modules.loss import _WeightedLoss
from torch_scatter import scatter_mean, scatter_add


def split_tensor_by_batch(x, batch, num_graphs=None):
    """
    Args:
        x:      (N, ...)
        batch:  (B, )
    Returns:
        [(N_1, ), (N_2, ) ..., (N_B, ))]
    """
    if num_graphs is None:
        num_graphs = batch.max().item() + 1
    x_split = []
    for i in range (num_graphs):
        mask = batch == i
        x_split.append(x[mask])
    return x_split


def concat_tensors_to_batch(x_split):
    x = torch.cat(x_split, dim=0)
    batch = torch.repeat_interleave(
        torch.arange(len(x_split)), 
        repeats=torch.LongTensor([s.size(0) for s in x_split])
    ).to(device=x.device)
    return x, batch


def split_tensor_to_segments(x, segsize):
    num_segs = math.ceil(x.size(0) / segsize)
    segs = []
    for i in range(num_segs):
        segs.append(x[i*segsize : (i+1)*segsize])
    return segs


def split_tensor_by_lengths(x, lengths):
    segs = []
    for l in lengths:
        segs.append(x[:l])
        x = x[l:]
    return segs


def batch_intersection_mask(batch, batch_filter):
    batch_filter = batch_filter.unique()
    mask = (batch.view(-1, 1) == batch_filter.view(1, -1)).any(dim=1)
    return mask


class MeanReadout(nn.Module):
    """Mean readout operator over graphs with variadic sizes."""

    def forward(self, input, batch, num_graphs):
        """
        Perform readout over the graph(s).
        Parameters:
            data (torch_geometric.data.Data): batched graph
            input (Tensor): node representations
        Returns:
            Tensor: graph representations
        """
        output = scatter_mean(input, batch, dim=0, dim_size=num_graphs)
        return output


class SumReadout(nn.Module):
    """Sum readout operator over graphs with variadic sizes."""

    def forward(self, input, batch, num_graphs):
        """
        Perform readout over the graph(s).
        Parameters:
            data (torch_geometric.data.Data): batched graph
            input (Tensor): node representations
        Returns:
            Tensor: graph representations
        """
        output = scatter_add(input, batch, dim=0, dim_size=num_graphs)
        return output


class MultiLayerPerceptron(nn.Module):
    """
    Multi-layer Perceptron.
    Note there is no activation or dropout in the last layer.
    Parameters:
        input_dim (int): input dimension
        hidden_dim (list of int): hidden dimensions
        activation (str or function, optional): activation function
        dropout (float, optional): dropout rate
    """

    def __init__(self, input_dim, hidden_dims, activation="relu", dropout=0):
        super(MultiLayerPerceptron, self).__init__()

        self.dims = [input_dim] + hidden_dims
        if isinstance(activation, str):
            self.activation = getattr(F, activation)
        else:
            self.activation = None
        if dropout:
            self.dropout = nn.Dropout(dropout)
        else:
            self.dropout = None

        self.layers = nn.ModuleList()
        for i in range(len(self.dims) - 1):
            self.layers.append(nn.Linear(self.dims[i], self.dims[i + 1]))

    def forward(self, input):
        """"""
        x = input
        for i, layer in enumerate(self.layers):
            x = layer(x)
            if i < len(self.layers) - 1:
                if self.activation:
                    x = self.activation(x)
                if self.dropout:
                    x = self.dropout(x)
        return x


class SmoothCrossEntropyLoss(_WeightedLoss):
    def __init__(self, weight=None, reduction='mean', smoothing=0.0):
        super().__init__(weight=weight, reduction=reduction)
        self.smoothing = smoothing
        self.weight = weight
        self.reduction = reduction

    @staticmethod
    def _smooth_one_hot(targets:torch.Tensor, n_classes:int, smoothing=0.0):
        assert 0 <= smoothing < 1
        with torch.no_grad():
            targets = torch.empty(size=(targets.size(0), n_classes),
                    device=targets.device) \
                .fill_(smoothing /(n_classes-1)) \
                .scatter_(1, targets.data.unsqueeze(1), 1.-smoothing)
        return targets

    def forward(self, inputs, targets):
        targets = SmoothCrossEntropyLoss._smooth_one_hot(targets, inputs.size(-1),
            self.smoothing)
        lsm = F.log_softmax(inputs, -1)

        if self.weight is not None:
            lsm = lsm * self.weight.unsqueeze(0)

        loss = -(targets * lsm).sum(-1)

        if  self.reduction == 'sum':
            loss = loss.sum()
        elif  self.reduction == 'mean':
            loss = loss.mean()

        return loss


class GaussianSmearing(nn.Module):
    def __init__(self, start=0.0, stop=10.0, num_gaussians=50):
        super().__init__()
        offset = torch.linspace(start, stop, num_gaussians)
        self.coeff = -0.5 / (offset[1] - offset[0]).item()**2
        self.register_buffer('offset', offset)

    def forward(self, dist):
        dist = dist.view(-1, 1) - self.offset.view(1, -1)
        return torch.exp(self.coeff * torch.pow(dist, 2))


class ShiftedSoftplus(nn.Module):
    def __init__(self):
        super().__init__()
        self.shift = torch.log(torch.tensor(2.0)).item()

    def forward(self, x):
        return F.softplus(x) - self.shift


def compose_context(h_protein, h_ligand, pos_protein, pos_ligand, batch_protein, batch_ligand):
    batch_ctx = torch.cat([batch_protein, batch_ligand], dim=0)
    sort_idx = batch_ctx.argsort()

    mask_protein = torch.cat([
        torch.ones([batch_protein.size(0)], device=batch_protein.device).bool(),
        torch.zeros([batch_ligand.size(0)], device=batch_ligand.device).bool(),
    ], dim=0)[sort_idx]

    batch_ctx = batch_ctx[sort_idx]
    h_ctx = torch.cat([h_protein, h_ligand], dim=0)[sort_idx]       # (N_protein+N_ligand, H)
    pos_ctx = torch.cat([pos_protein, pos_ligand], dim=0)[sort_idx] # (N_protein+N_ligand, 3)

    return h_ctx, pos_ctx, batch_ctx


def get_complete_graph(batch):
    """
    Args:
        batch:  Batch index.
    Returns:
        edge_index: (2, N_1 + N_2 + ... + N_{B-1}), where N_i is the number of nodes of the i-th graph.
        neighbors:  (B, ), number of edges per graph.
    """
    natoms = scatter_add(torch.ones_like(batch), index=batch, dim=0)

    natoms_sqr = (natoms ** 2).long()
    num_atom_pairs = torch.sum(natoms_sqr)
    natoms_expand = torch.repeat_interleave(natoms, natoms_sqr)

    index_offset = torch.cumsum(natoms, dim=0) - natoms
    index_offset_expand = torch.repeat_interleave(index_offset, natoms_sqr)

    index_sqr_offset = torch.cumsum(natoms_sqr, dim=0) - natoms_sqr
    index_sqr_offset = torch.repeat_interleave(index_sqr_offset, natoms_sqr)

    atom_count_sqr = torch.arange(num_atom_pairs, device=num_atom_pairs.device) - index_sqr_offset

    index1 = (atom_count_sqr // natoms_expand).long() + index_offset_expand
    index2 = (atom_count_sqr % natoms_expand).long() + index_offset_expand
    edge_index = torch.cat([index1.view(1, -1), index2.view(1, -1)])
    mask = torch.logical_not(index1 == index2)
    edge_index = edge_index[:, mask]

    num_edges = natoms_sqr - natoms # Number of edges per graph

    return edge_index, num_edges


def compose_context_stable(h_protein, h_ligand, pos_protein, pos_ligand, batch_protein, batch_ligand):
    num_graphs = batch_protein.max().item() + 1

    batch_ctx = []
    h_ctx = []
    pos_ctx = []
    mask_protein = []

    for i in range(num_graphs):
        mask_p, mask_l = (batch_protein == i), (batch_ligand == i)
        batch_p, batch_l = batch_protein[mask_p], batch_ligand[mask_l]

        batch_ctx += [batch_p, batch_l]
        h_ctx += [h_protein[mask_p], h_ligand[mask_l]]
        pos_ctx += [pos_protein[mask_p], pos_ligand[mask_l]]
        mask_protein += [
            torch.ones([batch_p.size(0)], device=batch_p.device, dtype=torch.bool),
            torch.zeros([batch_l.size(0)], device=batch_l.device, dtype=torch.bool),
        ]

    batch_ctx = torch.cat(batch_ctx, dim=0)
    h_ctx = torch.cat(h_ctx, dim=0)
    pos_ctx = torch.cat(pos_ctx, dim=0)
    mask_protein = torch.cat(mask_protein, dim=0)

    return h_ctx, pos_ctx, batch_ctx, mask_protein

if __name__ == '__main__':
    h_protein = torch.randn([60, 64])
    h_ligand = -torch.randn([33, 64])
    pos_protein = torch.clamp(torch.randn([60, 3]), 0, float('inf'))
    pos_ligand = torch.clamp(torch.randn([33, 3]), float('-inf'), 0)
    batch_protein = torch.LongTensor([0]*10 + [1]*20 + [2]*30)
    batch_ligand = torch.LongTensor([0]*11 + [1]*11 + [2]*11)

    h_ctx, pos_ctx, batch_ctx, mask_protein = compose_context_stable(h_protein, h_ligand, pos_protein, pos_ligand, batch_protein, batch_ligand)

    assert (batch_ctx[mask_protein] == batch_protein).all()
    assert (batch_ctx[torch.logical_not(mask_protein)] == batch_ligand).all()

    assert torch.allclose(h_ctx[torch.logical_not(mask_protein)], h_ligand)
    assert torch.allclose(h_ctx[mask_protein], h_protein)

    assert torch.allclose(pos_ctx[torch.logical_not(mask_protein)], pos_ligand)
    assert torch.allclose(pos_ctx[mask_protein], pos_protein)