id must be int64? about graph-learn

[archwalker]

Yes, id should be in the format of int64, in you case, you may preprocess the data to remap the card number to cardinal number.

[chenshthu]

thanks.
By the way, is the accuracy lost in the calculation of ID? During the call process, when the ID is 6221888200042590280, with the query result， Id becomes 6221888200042590208. There are other examples. The last 3 or 4 bits have changed. Please confirm

For example, a method is defined as

def node_ embedding(self, type='train'):
ego_ flow = gl.EgoFlow (self._ val_ sample_ seed,
self._ positive_ sample,
self._ receptive_ FN,
self.src_ ego_ spec)
iterator = ego_ flow.iterator
ego_ tensor = ego_ flow.pos_ src_ ego_ tensor
emb = self.encoders ['src'].encode(ego_ tensor)
ids = ego_ tensor.src.ids
return ids, emb, iterator

Call method

embs = trainer.get_ node_ embedding()
with tf.Session () as sess:
result = sess.run ( tf.math.argmax (embs[:,1:], 1, output_ type= tf.int32 )
with open("./result", 'w') as f:
for i in range( embs.shape [0]):
f.write(str(int(embs[i][0])) + "\t" + str(result[i]) + "\n")