sse_batch.py

"""
Learning Steady-States of Iterative Algorithms over Graphs
Paper: http://proceedings.mlr.press/v80/dai18a.html

"""
import argparse
import numpy as np
import time
import math
import mxnet as mx
from mxnet import gluon
import dgl
import dgl.function as fn
from dgl import DGLGraph
from dgl.data import register_data_args, load_data

def gcn_msg(edges):
    # TODO should we use concat?
    return {'m': mx.nd.concat(edges.src['in'], edges.src['h'], dim=1)}

def gcn_reduce(nodes):
    return {'accum': mx.nd.sum(nodes.mailbox['m'], 1) / nodes.mailbox['m'].shape[1]}

class NodeUpdate(gluon.Block):
    def __init__(self, out_feats, activation=None, alpha=0.1, **kwargs):
        super(NodeUpdate, self).__init__(**kwargs)
        self.linear1 = gluon.nn.Dense(out_feats, activation=activation)
        # TODO what is the dimension here?
        self.linear2 = gluon.nn.Dense(out_feats)
        self.alpha = alpha

    def forward(self, in_data, hidden_data, accum):
        tmp = mx.nd.concat(in_data, accum, dim=1)
        hidden = self.linear2(self.linear1(tmp))
        return hidden_data * (1 - self.alpha) + self.alpha * hidden

class DGLNodeUpdate(gluon.Block):
    def __init__(self, update):
        super(DGLNodeUpdate, self).__init__()
        self.update = update

    def forward(self, node):
        return {'h1': self.update(node.data['in'], node.data['h'], node.data['accum'])}

class SSEUpdateHidden(gluon.Block):
    def __init__(self,
                 n_hidden,
                 dropout,
                 activation,
                 **kwargs):
        super(SSEUpdateHidden, self).__init__(**kwargs)
        with self.name_scope():
            self.layer = NodeUpdate(n_hidden, activation)
        self.dropout = dropout

    def forward(self, g, vertices):
        if vertices is None:
            deg = mx.nd.expand_dims(g.in_degrees(np.arange(0, g.number_of_nodes())), 1).astype(np.float32)
            feat = g.get_n_repr()['in']
            cat = mx.nd.concat(feat, g.ndata['h'], dim=1)
            accum = mx.nd.dot(g.adjacency_matrix(), cat) / deg
            return self.layer(feat, g.ndata['h'], accum)
        else:
            deg = mx.nd.expand_dims(g.in_degrees(vertices), 1).astype(np.float32)
            # We don't need dropout for inference.
            if self.dropout:
                # TODO here we apply dropout on all vertex representation.
                g.ndata['h'] = mx.nd.Dropout(g.ndata['h'], p=self.dropout)
            feat = g.get_n_repr()['in']
            cat = mx.nd.concat(feat, g.ndata['h'], dim=1)
            slices = mx.nd.take(g.adjacency_matrix(), vertices).as_in_context(cat.context)
            accum = mx.nd.dot(slices, cat) / deg.as_in_context(cat.context)
            vertices = vertices.as_in_context(g.ndata['in'].context)
            return self.layer(mx.nd.take(feat, vertices),
                              mx.nd.take(g.ndata['h'], vertices), accum)

class DGLSSEUpdateHidden(gluon.Block):
    def __init__(self,
                 n_hidden,
                 activation,
                 dropout,
                 use_spmv,
                 **kwargs):
        super(DGLSSEUpdateHidden, self).__init__(**kwargs)
        with self.name_scope():
            self.layer = DGLNodeUpdate(NodeUpdate(n_hidden, activation))
        self.dropout = dropout
        self.use_spmv = use_spmv

    def forward(self, g, vertices):
        if self.use_spmv:
            feat = g.ndata['in']
            g.ndata['cat'] = mx.nd.concat(feat, g.ndata['h'], dim=1)

            msg_func = fn.copy_src(src='cat', out='m')
            reduce_func = fn.sum(msg='m', out='accum')
        else:
            msg_func = gcn_msg
            reduce_func = gcn_reduce
        deg = mx.nd.expand_dims(g.in_degrees(np.arange(0, g.number_of_nodes())), 1).astype(np.float32)
        if vertices is None:
            g.update_all(msg_func, reduce_func, None)
            if self.use_spmv:
                g.ndata.pop('cat')
                g.ndata['accum'] = g.ndata['accum'] / deg
            batch_size = 100000
            num_batches = int(math.ceil(g.number_of_nodes() / batch_size))
            for i in range(num_batches):
                vs = mx.nd.arange(i * batch_size, min((i + 1) * batch_size, g.number_of_nodes()), dtype=np.int64)
                g.apply_nodes(self.layer, vs, inplace=True)
            g.ndata.pop('accum')
            return g.get_n_repr()['h1']
        else:
            # We don't need dropout for inference.
            if self.dropout:
                # TODO here we apply dropout on all vertex representation.
                g.ndata['h'] = mx.nd.Dropout(g.ndata['h'], p=self.dropout)
            g.pull(vertices, msg_func, reduce_func, None)
            if self.use_spmv:
                g.ndata.pop('cat')
                deg = deg.as_in_context(g.ndata['accum'].context)
                g.ndata['accum'] = g.ndata['accum'] / deg
            g.apply_nodes(self.layer, vertices)
            g.ndata.pop('accum')
            return g.ndata['h1'][vertices.as_in_context(g.ndata['h1'].context)]

class SSEPredict(gluon.Block):
    def __init__(self, update_hidden, out_feats, dropout, **kwargs):
        super(SSEPredict, self).__init__(**kwargs)
        with self.name_scope():
            self.linear1 = gluon.nn.Dense(out_feats, activation='relu')
            self.linear2 = gluon.nn.Dense(out_feats)
        self.update_hidden = update_hidden
        self.dropout = dropout

    def forward(self, g, vertices):
        hidden = self.update_hidden(g, vertices)
        if self.dropout:
            hidden = mx.nd.Dropout(hidden, p=self.dropout)
        return self.linear2(self.linear1(hidden))

def copy_to_gpu(subg, ctx):
    frame = subg.ndata
    for key in frame:
        subg.ndata[key] = frame[key].as_in_context(ctx)

def main(args, data):
    if isinstance(data.features, mx.nd.NDArray):
        features = data.features
    else:
        features = mx.nd.array(data.features)
    if isinstance(data.labels, mx.nd.NDArray):
        labels = data.labels
    else:
        labels = mx.nd.array(data.labels)
    train_size = len(labels) * args.train_percent
    train_vs = np.arange(train_size, dtype='int64')
    eval_vs = np.arange(train_size, len(labels), dtype='int64')
    print("train size: " + str(len(train_vs)))
    print("eval size: " + str(len(eval_vs)))
    eval_labels = mx.nd.array(data.labels[eval_vs])
    in_feats = features.shape[1]
    n_classes = data.num_labels
    n_edges = data.graph.number_of_edges()

    # create the SSE model
    try:
        graph = data.graph.get_graph()
    except AttributeError:
        graph = data.graph
    g = DGLGraph(graph, readonly=True)
    g.ndata['in'] = features
    g.ndata['h'] = mx.nd.random.normal(shape=(g.number_of_nodes(), args.n_hidden),
            ctx=mx.cpu(0))

    update_hidden_infer = DGLSSEUpdateHidden(args.n_hidden, 'relu',
                                             args.update_dropout, args.use_spmv,
                                             prefix='sse')
    update_hidden_train = DGLSSEUpdateHidden(args.n_hidden, 'relu',
                                             args.update_dropout, args.use_spmv,
                                             prefix='sse')
    if not args.dgl:
        update_hidden_infer = SSEUpdateHidden(args.n_hidden, args.update_dropout, 'relu',
                                              prefix='sse')
        update_hidden_train = SSEUpdateHidden(args.n_hidden, args.update_dropout, 'relu',
                                              prefix='sse')

    model_train = SSEPredict(update_hidden_train, args.n_hidden, args.predict_dropout, prefix='app')
    model_infer = SSEPredict(update_hidden_infer, args.n_hidden, args.predict_dropout, prefix='app')
    model_infer.initialize(ctx=mx.cpu(0))
    if args.gpu <= 0:
        model_train.initialize(ctx=mx.cpu(0))
    else:
        train_ctxs = []
        for i in range(args.gpu):
            train_ctxs.append(mx.gpu(i))
        model_train.initialize(ctx=train_ctxs)

    # use optimizer
    num_batches = int(g.number_of_nodes() / args.batch_size)
    scheduler = mx.lr_scheduler.CosineScheduler(args.n_epochs * num_batches,
            args.lr * 10, 0, 0, args.lr/5)
    trainer = gluon.Trainer(model_train.collect_params(), 'adam', {'learning_rate': args.lr,
        'lr_scheduler': scheduler}, kvstore=mx.kv.create('device'))

    # compute vertex embedding.
    all_hidden = update_hidden_infer(g, None)
    g.ndata['h'] = all_hidden
    rets = []
    rets.append(all_hidden)

    # initialize graph
    dur = []
    for epoch in range(args.n_epochs):
        t0 = time.time()
        train_loss = 0
        i = 0
        num_batches = len(train_vs) / args.batch_size
        start1 = time.time()
        for subg, seeds in dgl.contrib.sampling.NeighborSampler(g, args.batch_size, g.number_of_nodes(),
                neighbor_type='in', num_workers=args.num_parallel_subgraphs, seed_nodes=train_vs,
                shuffle=True):
            subg.copy_from_parent()

            losses = []
            if args.gpu > 0:
                ctx = mx.gpu(i % args.gpu)
                copy_to_gpu(subg, ctx)

            subg_seeds = subg.map_to_subgraph_nid(seeds)
            with mx.autograd.record():
                logits = model_train(subg, subg_seeds.tousertensor())
                batch_labels = mx.nd.array(labels[seeds.asnumpy()], ctx=logits.context)
                loss = mx.nd.softmax_cross_entropy(logits, batch_labels)
            loss.backward()
            losses.append(loss)
            i += 1
            if args.gpu <= 0:
                trainer.step(seeds.shape[0])
                train_loss += loss.asnumpy()[0]
                losses = []
            elif i % args.gpu == 0:
                trainer.step(len(seeds) * len(losses))
                for loss in losses:
                    train_loss += loss.asnumpy()[0]
                losses = []

            if i % args.num_parallel_subgraphs == 0:
                end1 = time.time()
                print("process " + str(args.num_parallel_subgraphs)
                        + " subgraphs takes " + str(end1 - start1))
                start1 = end1

            if i > num_batches / 3:
                break

        # prediction.
        logits = model_infer(g, mx.nd.array(eval_vs, dtype=np.int64))
        eval_loss = mx.nd.softmax_cross_entropy(logits, eval_labels)
        eval_loss = eval_loss.asnumpy()[0]

        # update the inference model.
        infer_params = model_infer.collect_params()
        for key in infer_params:
            idx = trainer._param2idx[key]
            trainer._kvstore.pull(idx, out=infer_params[key].data())

        # Update node embeddings.
        all_hidden = update_hidden_infer(g, None)
        g.ndata['h'] = all_hidden
        rets.append(all_hidden)

        dur.append(time.time() - t0)
        print("Epoch {:05d} | Train Loss {:.4f} | Eval Loss {:.4f} | Time(s) {:.4f} | ETputs(KTEPS) {:.2f}".format(
            epoch, train_loss, eval_loss, np.mean(dur), n_edges / np.mean(dur) / 1000))

    return rets

class MXNetGraph(object):
    """A simple graph object that uses scipy matrix."""
    def __init__(self, mat):
        self._mat = mat

    def get_graph(self):
        return self._mat

    def number_of_nodes(self):
        return self._mat.shape[0]

    def number_of_edges(self):
        return mx.nd.contrib.getnnz(self._mat).asnumpy()[0]

class GraphData:
    def __init__(self, csr, num_feats):
        num_edges = mx.nd.contrib.getnnz(csr).asnumpy()[0]
        edge_ids = mx.nd.arange(0, num_edges, step=1, repeat=1, dtype=np.int64)
        csr = mx.nd.sparse.csr_matrix((edge_ids, csr.indices, csr.indptr), shape=csr.shape, dtype=np.int64)
        self.graph = MXNetGraph(csr)
        self.features = mx.nd.random.normal(shape=(csr.shape[0], num_feats))
        self.labels = mx.nd.floor(mx.nd.random.normal(loc=0, scale=10, shape=(csr.shape[0])))
        self.num_labels = 10

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='GCN')
    register_data_args(parser)
    parser.add_argument("--graph-file", type=str, default="",
            help="graph file")
    parser.add_argument("--num-feats", type=int, default=10,
            help="the number of features")
    parser.add_argument("--gpu", type=int, default=-1,
            help="gpu")
    parser.add_argument("--lr", type=float, default=1e-3,
            help="learning rate")
    parser.add_argument("--batch-size", type=int, default=128,
            help="number of vertices in a batch")
    parser.add_argument("--n-epochs", type=int, default=20,
            help="number of training epochs")
    parser.add_argument("--n-hidden", type=int, default=16,
            help="number of hidden gcn units")
    parser.add_argument("--warmup", type=int, default=10,
            help="number of iterations to warm up with large learning rate")
    parser.add_argument("--update-dropout", type=float, default=0,
            help="the dropout rate for updating vertex embedding")
    parser.add_argument("--predict-dropout", type=float, default=0,
            help="the dropout rate for prediction")
    parser.add_argument("--train_percent", type=float, default=0.5,
            help="the percentage of data used for training")
    parser.add_argument("--use-spmv", action="store_true",
            help="use SpMV for faster speed.")
    parser.add_argument("--dgl", action="store_true")
    parser.add_argument("--num-parallel-subgraphs", type=int, default=1,
            help="the number of subgraphs to construct in parallel.")
    args = parser.parse_args()

    # load and preprocess dataset
    if args.graph_file != '':
        csr = mx.nd.load(args.graph_file)[0]
        data = GraphData(csr, args.num_feats)
        csr = None
    else:
        data = load_data(args)
    rets1 = main(args, data)
    rets2 = main(args, data)
    for hidden1, hidden2 in zip(rets1, rets2):
        print("hidden: " + str(mx.nd.sum(mx.nd.abs(hidden1 - hidden2)).asnumpy()))