colflip / gnns_fewshot

code implementation of GNNs in few-shot learning: GCN, GAT, GraphSAGE to the node classification task.

gnns_fewshot [address]

Code implementation of GNNs in few-shot learning: GCN, GAT, GraphSAGE to the node classification task of some datasets.

dependencies

• Python ≥ 3.10
• PyTorch ≥ 11.3
• pyg ≥ 1.12.0

results

novel class is num_class*2/5

	model/shot/dataset	Cora	CiteSeer	Photo	cs	Computers	CoraFull
GCN	1	72.43±1.97[2]	64.9±1.75[2]	78.69±2.49[3]	83.51±0.88[6]	74.35±1.99[4]	34.56±0.22[28]
	3	85.6±1.13[2]	75.67±1.66[2]	91.12±0.54[3]	91.92±0.2[6]	87.88±0.74[4]	54.92±0.17[28]
	5	89.28±0.8[2]	79.29±1.39[2]	93.32±0.32[3]	93.75±0.14[6]	90.64±0.46[4]	62.46±0.15[28]
GAT	1	69.37±2.34[2]	61.08±1.59[2]	40.52±3.98[3]	73.39±1.53[6]	30.95±3.1[4]	34.64±0.26[28]
	3	84.07±1.31[2]	72.12±1.87[2]	61.33±8.15[3]	89.95±0.28[6]	56.33±7.8[4]	54.55±0.17[28]
	5	88.79±0.9[2]	77.12±1.57[2]	74.29±7.08[3]	92.17±0.17[6]	69.63±7.27[4]	62.05±0.15[28]
GraphSAGE	1	71.74±1.75[2]	64.41±1.78[2]	48.61±1.77[3]	72.7±2.07[6]	36.4±0.91[4]	23.67±0.26[28]
	3	83.32±1.15[2]	72.98±1.59[2]	69.95±2.03[3]	87.24±0.64[6]	62.1±1.7[4]	47.93±0.18[28]
	5	87.48±0.92[2]	78.41±1.25[2]	82.17±1.13[3]	90.6±0.31[6]	75.69±1.31[4]	57.68±0.16[28]

novel class is 2

	model/shot/dataset	Cora	CiteSeer	Photo	cs	Computers	CoraFull	PubMed
GCN	1	72.12±1.96	64.82±1.76	83.82±2.76	94.6±0.86	83.53±3.21
	3	85.62±1.09	75.43±1.73	93.87±0.58	97.86±0.17	93.9±0.94
	5	89.18±0.77	79.28±1.41	95.38±0.31	98.08±0.15	95.56±0.43
GAT	1	68.66±2.31	61.27±1.58	56.78±4.56	86.42±2.27	51.03±3.78
	3	84.22±1.3	71.94±1.84	68.62±7.07	95.57±0.47	65.88±7.03
	5	88.92±0.92	77.54±1.53	76.12±6.48	96.82±0.27	75.34±7.65
GraphSAGE	1	71.38±1.78	64.55±1.64	60.6±1.82	89.86±1.79	57.57±1.32
	3	83.63±1.11	73.38±1.57	76.25±2.6	96.03±0.5	69.92±2.49
	5	87.7±0.84	77.54±1.27	84.35±1.58	96.3±0.33	80.01±2.3

license

It is under the MIT license. See the LICENSE file for details.

---

“一切恩爱会、无常难得久、生世多畏惧、命危于晨露，由爱故生忧，由爱故生怖，若离于爱者，无忧亦无怖。”