upload necessary data and preprocess code

.gitignore

@@ -0,0 +1,9 @@
+lp/data/
+lp/preprocessed/Amazon/unconnected_pairs_offset.npy
+lp/preprocessed/Amazon/neg_ratings_offset.npy
+lp/preprocessed/Douban_Movie/neg_ratings_offset.npy
+lp/preprocessed/Douban_Movie/unconnected_pairs_offset.npy
+lp/preprocessed/Yelp/neg_ratings_offset.npy
+lp/preprocessed/Yelp/unconnected_pairs_offset.npy
+
+nc/data/

lp/gen_neg.py

@@ -0,0 +1,40 @@
+import numpy as np
+import scipy.sparse as sp
+import os
+import torch
+import sys
+
+def main(prefix):
+
+    pos_pairs_offset = np.load(os.path.join(prefix, "pos_pairs_offset.npz"))
+    unconnected_pairs_offset = np.load(os.path.join(prefix, "unconnected_pairs_offset.npy"))
+    neg_ratings_offset = np.load(os.path.join(prefix, "neg_ratings_offset.npy"))
+
+    train_len = pos_pairs_offset['train'].shape[0]
+    val_len = pos_pairs_offset['val'].shape[0]
+    test_len = pos_pairs_offset['test'].shape[0]
+    pos_len = train_len + val_len + test_len
+
+    if pos_len > neg_ratings_offset.shape[0]:
+        indices = np.arange(unconnected_pairs_offset.shape[0])
+        assert(indices.shape[0] > pos_len)
+        np.random.shuffle(indices)
+        makeup = indices[:pos_len - neg_ratings_offset.shape[0]]
+        neg_ratings_offset = np.concatenate((neg_ratings_offset, unconnected_pairs_offset[makeup]), axis=0)
+        assert(pos_len == neg_ratings_offset.shape[0])
+
+    indices = np.arange(neg_ratings_offset.shape[0])
+    np.random.shuffle(indices)
+    np.savez(os.path.join(prefix, "neg_pairs_offset"), train=neg_ratings_offset[indices[:train_len]],
+                                                        val=neg_ratings_offset[indices[train_len:train_len + val_len]],
+                                                        test=neg_ratings_offset[indices[train_len + val_len:pos_len]])
+
+if __name__ == '__main__':
+    dataset = sys.argv[1]
+    prefix = os.path.join("./preprocessed/", dataset)
+    np.random.seed(int(sys.argv[2]))
+    main(prefix)
+
+    #! Yelp 2
+    #! Amazon 4
+    #! Douban_Movie 6

nc/preprocess.py

@@ -0,0 +1,36 @@
+import os
+import sys
+import numpy as np
+import torch
+import pickle as pkl
+
+cstr_nc = {
+    "DBLP" : [1, 4],
+    "ACM" : [0, 2, 4],
+    "IMDB" : [0, 2, 4]
+}
+
+def main(dataset):
+    prefix = os.path.join("./data/", dataset)
+    with open(os.path.join(prefix, "edges.pkl"), "rb") as f:
+        edges = pkl.load(f)
+        f.close()
+
+    node_types = np.zeros((edges[0].shape[0],), dtype=np.int32)
+
+    a = np.unique(list(edges[0].tocoo().row) + list(edges[2].tocoo().row))
+    b = np.unique(edges[0].tocoo().col)
+    c = np.unique(edges[2].tocoo().col)
+    print(a.shape[0], b.shape[0], c.shape[0])
+    assert(a.shape[0] + b.shape[0] + c.shape[0] == node_types.shape[0])
+    assert(np.unique(np.concatenate((a, b, c))).shape[0] == node_types.shape[0])
+
+    node_types[a.shape[0]:a.shape[0] + b.shape[0]] = 1
+    node_types[a.shape[0] + b.shape[0]:] = 2
+    assert(node_types.sum() == b.shape[0] + 2 * c.shape[0])
+    np.save(os.path.join(prefix, "node_types"), node_types)
+
+if __name__ == "__main__":
+    main("DBLP")
+    main("ACM")
+    main("IMDB")