From f62b619094b586a0d73047351b08d453f1e136c3 Mon Sep 17 00:00:00 2001 From: stevenHust Date: Wed, 6 Sep 2023 14:35:29 +0800 Subject: [PATCH] Update README.md (#4) * Update README.md * Update ReadMe.md * Update README.md * Update data_prepare.py * Update readme.md * Update README.md * Update readme.md * Update ReadMe.md * Update ReadMe.md * Update ReadMe.md * Update readme.md * Update ReadMe.md * Update data_preprcoss.py * Update run_ssr_link.sh * Update ReadMe.md * Update readme.md * Update readme.md * Update readme.md * Update readme.md * Update readme.md * Update readme.md * fix directory bug * tiny fix for nasa * revise install * Update unit_testing.yml --------- Co-authored-by: dalong.zdl --- .github/workflows/unit_testing.yml | 2 +- agl/python/examples/drgst/README.md | 4 +- agl/python/examples/geniepath_ppi/ReadMe.md | 2 +- .../data_process/data_prepare.py | 2 +- agl/python/examples/hegnn_acm/ReadMe.md | 5 +- agl/python/examples/kcan_movielens/readme.md | 19 +- agl/python/examples/nasa/README.md | 308 +----------------- agl/python/examples/nasa/nasa.py | 14 +- agl/python/examples/pagnn/ReadMe.md | 69 +--- .../ssr_lastfm/data_process/data_preprcoss.py | 10 +- .../ssr_lastfm/data_process/run_ssr_link.sh | 12 +- agl/python/examples/ssr_lastfm/readme.md | 57 +--- .../examples/ssr_lastfm/ssr2_ego_eval.py | 2 +- .../examples/ssr_lastfm/ssr_ego_infer.py | 4 +- build.sh | 3 + doc/core/install.md | 16 +- 16 files changed, 74 insertions(+), 455 deletions(-) diff --git a/.github/workflows/unit_testing.yml b/.github/workflows/unit_testing.yml index d9ae7c0..d7d348c 100644 --- a/.github/workflows/unit_testing.yml +++ b/.github/workflows/unit_testing.yml @@ -18,4 +18,4 @@ jobs: docker run --net=host --rm -i -m 30000m -v ${GITHUB_WORKSPACE}:/graph_ml -w /graph_ml aglimage/agl:agl-ubuntu-gcc9.4.0-py3.8-cuda11.8-pytorch2.0.1-0825 /bin/bash -c 'cd agl/java && mvn -B package --file pom.xml' - name: python cpp unit testing if: always() - run: docker run --net=host --rm -i -m 30000m -v ${GITHUB_WORKSPACE}:/graph_ml -w /graph_ml aglimage/agl:agl-ubuntu-gcc9.4.0-py3.8-cuda11.8-pytorch2.0.1-0825 /bin/bash -c 'git config --global --add safe.directory /graph_ml && bash build.sh' + run: docker run --net=host --rm -i -m 30000m -v ${GITHUB_WORKSPACE}:/graph_ml -w /graph_ml aglimage/agl:agl-ubuntu-gcc9.4.0-py3.8-cuda11.8-pytorch2.0.1-0825 /bin/bash -c 'bash build.sh' diff --git a/agl/python/examples/drgst/README.md b/agl/python/examples/drgst/README.md index 7483863..8bbb0e3 100644 --- a/agl/python/examples/drgst/README.md +++ b/agl/python/examples/drgst/README.md @@ -20,12 +20,13 @@ stage = 6 ## Benchmark * 数据下载： - 从https://github.com/tkipf/gcn/tree/master/gcn/data下载ind.citeseer.开头的数据文件,放在data_process/data/目录下 + 从 https://github.com/tkipf/gcn/tree/master/gcn/data 下载ind.citeseer.开头的数据文件,放在data_process/data/目录下 * 数据预处理与子图采样： 运行submit.sh进行数据预处理和spark采样,得到训练集测试集验证集 * 模型 python drgst_citeseer.py * 效果 +``` In stage 0 test loss:0.9647, test acc:0.7070 In stage 1 @@ -38,3 +39,4 @@ In stage 4 test loss:0.9230, test acc:0.7440 In stage 5 test loss:0.9430, test acc:0.7540 +``` diff --git a/agl/python/examples/geniepath_ppi/ReadMe.md b/agl/python/examples/geniepath_ppi/ReadMe.md index d7bd8de..a8335ce 100644 --- a/agl/python/examples/geniepath_ppi/ReadMe.md +++ b/agl/python/examples/geniepath_ppi/ReadMe.md @@ -9,7 +9,7 @@ ## 说明 * 数据下载： - 从https://github.com/sufeidechabei/PPI-Inductive/tree/master/ppi下载,放到data_process/ppi/目录下 + 从 https://github.com/sufeidechabei/PPI-Inductive/tree/master/ppi 下载,放到data_process/ppi/目录下 * 数据预处理与子图采样： 运行submit.sh进行数据预处理和spark采样,得到训练集测试集验证集 * 模型 diff --git a/agl/python/examples/geniepath_ppi/data_process/data_prepare.py b/agl/python/examples/geniepath_ppi/data_process/data_prepare.py index 3fcd763..c4efc19 100644 --- a/agl/python/examples/geniepath_ppi/data_process/data_prepare.py +++ b/agl/python/examples/geniepath_ppi/data_process/data_prepare.py @@ -49,7 +49,7 @@ def load_data(prefix): if node["test"] and node["val"]: print("both test and val error:" + nid) exit(0) - train_graph_id = np.load(prefix + "-train_graph_id.npy") + train_graph_id = np.load("ppi/train_graph_id.npy") class_map = json.load(open(prefix + "-class_map.json")) with open('ppi_label.csv', 'w') as outfile: outfile.write('node_id,seed,label,train_flag\n') diff --git a/agl/python/examples/hegnn_acm/ReadMe.md b/agl/python/examples/hegnn_acm/ReadMe.md index 2b59476..cc7d16b 100644 --- a/agl/python/examples/hegnn_acm/ReadMe.md +++ b/agl/python/examples/hegnn_acm/ReadMe.md @@ -10,13 +10,13 @@ ### 数据 -下载数据https://drive.google.com/drive/folders/1koV0rGhj-UXrEMOCZezK1tnwC6zb69uB?usp=sharing ，将node.csv,edge.csv,label.csv文件拷贝到data_process目录下 +下载数据 https://drive.google.com/drive/folders/1koV0rGhj-UXrEMOCZezK1tnwC6zb69uB?usp=sharing ，将node.csv,edge.csv,label.csv文件拷贝到data_process目录下 ### 数据预处理与子图采样： 运行submit.sh进行spark采样,得到训练集测试集验证集 ### 效果 python model_hegnn.py - +``` Epoch: 01, Loss: 0.6549, val_micro_f1: 0.3533, test_micro_f1: 0.3812, time_cost:10.1865 (Epoch: 01, best_val_micro_f1: 0.3533, best_test_micro_f1: 0.3812)
Epoch: 02, Loss: 0.5937, val_micro_f1: 0.8100, test_micro_f1: 0.8640, time_cost:7.7812 @@ -218,3 +218,4 @@ Epoch: 99, Loss: 0.0003, val_micro_f1: 0.8767, test_micro_f1: 0.9068, time_cost: Epoch: 100, Loss: 0.0003, val_micro_f1: 0.8767, test_micro_f1: 0.9068, time_cost:7.9536 (Epoch: 07, best_val_micro_f1: 0.8900, best_test_micro_f1: 0.9144)
sucess +``` diff --git a/agl/python/examples/kcan_movielens/readme.md b/agl/python/examples/kcan_movielens/readme.md index 04e8c40..f94b648 100755 --- a/agl/python/examples/kcan_movielens/readme.md +++ b/agl/python/examples/kcan_movielens/readme.md @@ -14,14 +14,17 @@ ## 说明 > Warning: 模型部分实现和论文中略有不同，原文是知识图谱表征学习和kcan交替训练，这里只有kcan的训练。同时开源数据只有正边没有负边，所以负样本是随机采样的导致数据集中负样本和原论文不一样，因此效果并不能完全对齐论文。 -由于link类算法的样本数量过多，只能在分布式模式（比如yarn）运行，为了方便不想搭建yarn集群的用户，我们提供了采样好的图样本下载地址为:，用户可以直接进行模型训练 + ### 数据下载： -从https://drive.google.com/drive/folders/12_mU1jt7ntuWEMQ-bogF0cLQjFJijnab?usp=sharing下载数据文件,把图数据文件node_table.csv,link_table.csv,edge_table.csv放在data_process/目录下。 -由于link模式的样本量巨大，用户需要搭建spark集群运行。对于无法搭建集群的用户，可以下载预先采样的子图数据part-subgraph_kcan_train_test.csv，放在data_process/output_graph_feature目录下 +从 https://drive.google.com/drive/folders/12_mU1jt7ntuWEMQ-bogF0cLQjFJijnab?usp=sharing 下载数据文件,把图数据文件node_table.csv,link_table.csv,edge_table.csv放在data_process/目录下。 ### 数据预处理 以movielens为例子 -首先我们要把原始数据压缩成子图(pb string)的形式，使用如下data_process/submit.sh的命令 + +首先我们要把原始数据压缩成子图(pb string)的形式，使用如下data_process/submit.sh的命令。 + +由于link模式的样本量巨大，用户需要搭建spark集群运行。对于无法搭建集群的用户，可以从上面的链接中下载预先采样的子图数据part-subgraph_kcan_train_test.csv，放在data_process/output_graph_feature目录下 + ``` base=`dirname "$0"` cd "$base" @@ -50,7 +53,11 @@ python ../../run_spark.py \ - input_node_feature - 包含node_id,node_feature两个字段 -运行data_process/split_graph_features.py，将output_graph_feature目录下的子图根据train_flag划分为subgraph_kcan_movielens_train.txt和subgraph_kcan_movielens_test.txt文件给下游训练。 +``` +cd data_process +python split_graph_features.py +``` +运行上面的脚本，将output_graph_feature目录下的子图根据train_flag划分为subgraph_kcan_movielens_train.txt和subgraph_kcan_movielens_test.txt文件给下游训练。 ### 模型运行 ``` @@ -62,7 +69,7 @@ python kcan_subgraph_adj.py * kcan 未调参，100 epoch, AUC ~ 0.9 左右 (原论文 0.907) * 效率 * kcan - ``` +``` Epoch: 01, Loss: 0.4570, auc: 0.8826, best_auc: 0.8826, train_time: 154.442953, val_time: 30.473222 Epoch: 02, Loss: 0.4234, auc: 0.8841, best_auc: 0.8841, train_time: 151.677146, val_time: 29.973186 Epoch: 03, Loss: 0.4215, auc: 0.8842, best_auc: 0.8842, train_time: 153.905838, val_time: 32.188418 diff --git a/agl/python/examples/nasa/README.md b/agl/python/examples/nasa/README.md index 82d1f39..a2334d8 100644 --- a/agl/python/examples/nasa/README.md +++ b/agl/python/examples/nasa/README.md @@ -21,7 +21,7 @@ 以 `Amazon Photo` 数据集为例子 * 数据下载： - 从https://github.com/BUPT-GAMMA/NASA/tree/main/dataset下载photo.npz文件,放到data_process/dataset/目录下 + 从 https://github.com/BUPT-GAMMA/NASA/tree/main/dataset 下载photo.npz文件,放到data_process/dataset/目录下 * 数据预处理与子图采样： 运行submit.sh进行数据预处理和spark采样,得到每条样本的子图 将得到的csv表 `graph_features.csv`，包含 'seed', 'graph_feature', 'node_id_list', 'label_list', 'flag_list' 三个字段，其中 'flag_list' 字段中以 'train', 'val', 'test' 标记训练集数据、验证集数据和测试集数据。然后，执行下述指令，转换flag_list的格式以便于后续处理： @@ -38,308 +38,4 @@ python nasa.py ## BenchMark * 效果 - * NASA 未调参，300 epoch, Amazon Photo, F1 ~ 0.92 左右。 - -* 效率 * NASA - ``` - Epoch: 01, Loss: 0.7849, train_time: 2.409370, val_time: 0.387727 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.0901 - Epoch: 02, Loss: 1.9168, train_time: 0.463835, val_time: 0.409787 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.1148 - Epoch: 03, Loss: 1.3910, train_time: 0.461324, val_time: 0.389210 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.1148 - Epoch: 04, Loss: 1.3309, train_time: 0.450482, val_time: 0.384466 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.0390 - Epoch: 05, Loss: 1.0456, train_time: 0.454981, val_time: 0.381707 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.1190 - Epoch: 06, Loss: 0.9304, train_time: 0.449232, val_time: 0.385375 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.1196 - Epoch: 07, Loss: 0.7828, train_time: 0.462663, val_time: 0.385535 train_f1: 0.156250, val_f1: 0.1458, test_f1: 0.2350 - Epoch: 08, Loss: 0.7248, train_time: 0.450349, val_time: 0.388324 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2257 - Epoch: 09, Loss: 0.9742, train_time: 0.457985, val_time: 0.388683 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2257 - Epoch: 10, Loss: 0.9428, train_time: 0.446913, val_time: 0.387501 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2261 - Epoch: 11, Loss: 0.7487, train_time: 0.446201, val_time: 0.382103 train_f1: 0.225000, val_f1: 0.2250, test_f1: 0.4423 - Epoch: 12, Loss: 0.6091, train_time: 0.462892, val_time: 0.383698 train_f1: 0.131250, val_f1: 0.1250, test_f1: 0.2610 - Epoch: 13, Loss: 0.5671, train_time: 0.446771, val_time: 0.384869 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 14, Loss: 0.5563, train_time: 0.443244, val_time: 0.380741 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 15, Loss: 0.5612, train_time: 0.447747, val_time: 0.382180 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2611 - Epoch: 16, Loss: 0.5596, train_time: 0.446157, val_time: 0.382596 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2611 - Epoch: 17, Loss: 0.5488, train_time: 0.451101, val_time: 0.379544 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 18, Loss: 0.5509, train_time: 0.441869, val_time: 0.376832 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 19, Loss: 0.5207, train_time: 0.442937, val_time: 0.481601 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2610 - Epoch: 20, Loss: 0.5072, train_time: 0.558334, val_time: 0.377892 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 21, Loss: 0.5019, train_time: 0.467034, val_time: 0.402186 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 22, Loss: 0.4987, train_time: 0.457480, val_time: 0.390974 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2608 - Epoch: 23, Loss: 0.4962, train_time: 0.442622, val_time: 0.386423 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2612 - Epoch: 24, Loss: 0.4734, train_time: 0.462875, val_time: 0.384645 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2612 - Epoch: 25, Loss: 0.4778, train_time: 0.450900, val_time: 0.381567 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2611 - Epoch: 26, Loss: 0.4685, train_time: 0.449544, val_time: 0.380866 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2617 - Epoch: 27, Loss: 0.4658, train_time: 0.449051, val_time: 0.383625 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2621 - Epoch: 28, Loss: 0.4489, train_time: 0.450661, val_time: 0.387177 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2621 - Epoch: 29, Loss: 0.4529, train_time: 0.453003, val_time: 0.389676 train_f1: 0.125000, val_f1: 0.1250, test_f1: 0.2619 - Epoch: 30, Loss: 0.4527, train_time: 0.457725, val_time: 0.390348 train_f1: 0.137500, val_f1: 0.1292, test_f1: 0.2640 - Epoch: 31, Loss: 0.4319, train_time: 0.453672, val_time: 0.401276 train_f1: 0.137500, val_f1: 0.1333, test_f1: 0.2713 - Epoch: 32, Loss: 0.4339, train_time: 0.449958, val_time: 0.387022 train_f1: 0.156250, val_f1: 0.1458, test_f1: 0.2828 - Epoch: 33, Loss: 0.4262, train_time: 0.448586, val_time: 0.400396 train_f1: 0.150000, val_f1: 0.1333, test_f1: 0.2714 - Epoch: 34, Loss: 0.4131, train_time: 0.486054, val_time: 0.419609 train_f1: 0.143750, val_f1: 0.1292, test_f1: 0.2655 - Epoch: 35, Loss: 0.4172, train_time: 0.491612, val_time: 0.427876 train_f1: 0.156250, val_f1: 0.1417, test_f1: 0.2908 - Epoch: 36, Loss: 0.3963, train_time: 0.469271, val_time: 0.386648 train_f1: 0.181250, val_f1: 0.1542, test_f1: 0.2985 - Epoch: 37, Loss: 0.3875, train_time: 0.454999, val_time: 0.384635 train_f1: 0.206250, val_f1: 0.1542, test_f1: 0.3023 - Epoch: 38, Loss: 0.3801, train_time: 0.459137, val_time: 0.384290 train_f1: 0.212500, val_f1: 0.1542, test_f1: 0.3050 - Epoch: 39, Loss: 0.3738, train_time: 0.566178, val_time: 0.411133 train_f1: 0.250000, val_f1: 0.1667, test_f1: 0.3199 - Epoch: 40, Loss: 0.3738, train_time: 0.458307, val_time: 0.374921 train_f1: 0.287500, val_f1: 0.2000, test_f1: 0.3567 - Epoch: 41, Loss: 0.3691, train_time: 0.453641, val_time: 0.378376 train_f1: 0.350000, val_f1: 0.2833, test_f1: 0.3851 - Epoch: 42, Loss: 0.3576, train_time: 0.457642, val_time: 0.379780 train_f1: 0.412500, val_f1: 0.3417, test_f1: 0.4098 - Epoch: 43, Loss: 0.3432, train_time: 0.455808, val_time: 0.383137 train_f1: 0.468750, val_f1: 0.3833, test_f1: 0.4527 - Epoch: 44, Loss: 0.3314, train_time: 0.455915, val_time: 0.382015 train_f1: 0.481250, val_f1: 0.4333, test_f1: 0.4888 - Epoch: 45, Loss: 0.3308, train_time: 0.458668, val_time: 0.377257 train_f1: 0.487500, val_f1: 0.4292, test_f1: 0.4946 - Epoch: 46, Loss: 0.3190, train_time: 0.458364, val_time: 0.376608 train_f1: 0.487500, val_f1: 0.4167, test_f1: 0.4810 - Epoch: 47, Loss: 0.3213, train_time: 0.453923, val_time: 0.382437 train_f1: 0.493750, val_f1: 0.4292, test_f1: 0.4899 - Epoch: 48, Loss: 0.3125, train_time: 0.456032, val_time: 0.376015 train_f1: 0.512500, val_f1: 0.4500, test_f1: 0.5046 - Epoch: 49, Loss: 0.3140, train_time: 0.458869, val_time: 0.374605 train_f1: 0.506250, val_f1: 0.4542, test_f1: 0.5074 - Epoch: 50, Loss: 0.3123, train_time: 0.454955, val_time: 0.375020 train_f1: 0.562500, val_f1: 0.4875, test_f1: 0.5378 - Epoch: 51, Loss: 0.2811, train_time: 0.454685, val_time: 0.380710 train_f1: 0.606250, val_f1: 0.5292, test_f1: 0.5572 - Epoch: 52, Loss: 0.2915, train_time: 0.463989, val_time: 0.375225 train_f1: 0.612500, val_f1: 0.5542, test_f1: 0.5694 - Epoch: 53, Loss: 0.2807, train_time: 0.451378, val_time: 0.371463 train_f1: 0.618750, val_f1: 0.5625, test_f1: 0.5750 - Epoch: 54, Loss: 0.2847, train_time: 0.451757, val_time: 0.375171 train_f1: 0.631250, val_f1: 0.5792, test_f1: 0.5892 - Epoch: 55, Loss: 0.2789, train_time: 0.454146, val_time: 0.379980 train_f1: 0.650000, val_f1: 0.5833, test_f1: 0.6065 - Epoch: 56, Loss: 0.2717, train_time: 0.454165, val_time: 0.373986 train_f1: 0.643750, val_f1: 0.5875, test_f1: 0.6319 - Epoch: 57, Loss: 0.2691, train_time: 0.454717, val_time: 0.373815 train_f1: 0.650000, val_f1: 0.5833, test_f1: 0.6317 - Epoch: 58, Loss: 0.2592, train_time: 0.500981, val_time: 0.374715 train_f1: 0.650000, val_f1: 0.5875, test_f1: 0.6319 - Epoch: 59, Loss: 0.2554, train_time: 0.453532, val_time: 0.374376 train_f1: 0.650000, val_f1: 0.6167, test_f1: 0.6542 - Epoch: 60, Loss: 0.2652, train_time: 0.452625, val_time: 0.375079 train_f1: 0.668750, val_f1: 0.6250, test_f1: 0.6801 - Epoch: 61, Loss: 0.2515, train_time: 0.459745, val_time: 0.379440 train_f1: 0.693750, val_f1: 0.6583, test_f1: 0.7295 - Epoch: 62, Loss: 0.2585, train_time: 0.454022, val_time: 0.375827 train_f1: 0.712500, val_f1: 0.6833, test_f1: 0.7543 - Epoch: 63, Loss: 0.2483, train_time: 0.451904, val_time: 0.379126 train_f1: 0.725000, val_f1: 0.7042, test_f1: 0.7677 - Epoch: 64, Loss: 0.2423, train_time: 0.456659, val_time: 0.382272 train_f1: 0.718750, val_f1: 0.7000, test_f1: 0.7599 - Epoch: 65, Loss: 0.2576, train_time: 0.456727, val_time: 0.375665 train_f1: 0.718750, val_f1: 0.6917, test_f1: 0.7514 - Epoch: 66, Loss: 0.2429, train_time: 0.455046, val_time: 0.383324 train_f1: 0.725000, val_f1: 0.7125, test_f1: 0.7646 - Epoch: 67, Loss: 0.2394, train_time: 0.456770, val_time: 0.377453 train_f1: 0.756250, val_f1: 0.7292, test_f1: 0.7897 - Epoch: 68, Loss: 0.2421, train_time: 0.451637, val_time: 0.371791 train_f1: 0.731250, val_f1: 0.7208, test_f1: 0.7839 - Epoch: 69, Loss: 0.2441, train_time: 0.459214, val_time: 0.409385 train_f1: 0.712500, val_f1: 0.6917, test_f1: 0.7636 - Epoch: 70, Loss: 0.2277, train_time: 0.456779, val_time: 0.374898 train_f1: 0.700000, val_f1: 0.6833, test_f1: 0.7563 - Epoch: 71, Loss: 0.2288, train_time: 0.455686, val_time: 0.382077 train_f1: 0.712500, val_f1: 0.7125, test_f1: 0.7637 - Epoch: 72, Loss: 0.2276, train_time: 0.451888, val_time: 0.373402 train_f1: 0.731250, val_f1: 0.7333, test_f1: 0.7731 - Epoch: 73, Loss: 0.2343, train_time: 0.450120, val_time: 0.378067 train_f1: 0.737500, val_f1: 0.7250, test_f1: 0.7873 - Epoch: 74, Loss: 0.2235, train_time: 0.454213, val_time: 0.376312 train_f1: 0.737500, val_f1: 0.7333, test_f1: 0.7937 - Epoch: 75, Loss: 0.2164, train_time: 0.452079, val_time: 0.383587 train_f1: 0.731250, val_f1: 0.7250, test_f1: 0.7879 - Epoch: 76, Loss: 0.2290, train_time: 0.457238, val_time: 0.390302 train_f1: 0.737500, val_f1: 0.7417, test_f1: 0.7950 - Epoch: 77, Loss: 0.2215, train_time: 0.482541, val_time: 0.384275 train_f1: 0.743750, val_f1: 0.7458, test_f1: 0.7990 - Epoch: 78, Loss: 0.2212, train_time: 0.460082, val_time: 0.388165 train_f1: 0.750000, val_f1: 0.7500, test_f1: 0.7990 - Epoch: 79, Loss: 0.2171, train_time: 0.452800, val_time: 0.394711 train_f1: 0.743750, val_f1: 0.7375, test_f1: 0.7917 - Epoch: 80, Loss: 0.2178, train_time: 0.459813, val_time: 0.405177 train_f1: 0.743750, val_f1: 0.7292, test_f1: 0.7949 - Epoch: 81, Loss: 0.2080, train_time: 0.466736, val_time: 0.394859 train_f1: 0.750000, val_f1: 0.7375, test_f1: 0.7994 - Epoch: 82, Loss: 0.2187, train_time: 0.454182, val_time: 0.410645 train_f1: 0.731250, val_f1: 0.7458, test_f1: 0.8021 - Epoch: 83, Loss: 0.2125, train_time: 0.499220, val_time: 0.414889 train_f1: 0.731250, val_f1: 0.7292, test_f1: 0.7863 - Epoch: 84, Loss: 0.2119, train_time: 0.451876, val_time: 0.395117 train_f1: 0.725000, val_f1: 0.7208, test_f1: 0.7814 - Epoch: 85, Loss: 0.2101, train_time: 0.490785, val_time: 0.434433 train_f1: 0.743750, val_f1: 0.7375, test_f1: 0.7927 - Epoch: 86, Loss: 0.2152, train_time: 0.489830, val_time: 0.401396 train_f1: 0.787500, val_f1: 0.7583, test_f1: 0.8164 - Epoch: 87, Loss: 0.2115, train_time: 0.483145, val_time: 0.403199 train_f1: 0.781250, val_f1: 0.7833, test_f1: 0.8313 - Epoch: 88, Loss: 0.2150, train_time: 0.458329, val_time: 0.393530 train_f1: 0.756250, val_f1: 0.7542, test_f1: 0.8097 - Epoch: 89, Loss: 0.2079, train_time: 0.468917, val_time: 0.403194 train_f1: 0.743750, val_f1: 0.7333, test_f1: 0.7920 - Epoch: 90, Loss: 0.2136, train_time: 0.453062, val_time: 0.392831 train_f1: 0.800000, val_f1: 0.7792, test_f1: 0.8324 - Epoch: 91, Loss: 0.2048, train_time: 0.461432, val_time: 0.387603 train_f1: 0.850000, val_f1: 0.8042, test_f1: 0.8506 - Epoch: 92, Loss: 0.2057, train_time: 0.453992, val_time: 0.397924 train_f1: 0.856250, val_f1: 0.8000, test_f1: 0.8519 - Epoch: 93, Loss: 0.1995, train_time: 0.443586, val_time: 0.384226 train_f1: 0.818750, val_f1: 0.7875, test_f1: 0.8392 - Epoch: 94, Loss: 0.2098, train_time: 0.439300, val_time: 0.389837 train_f1: 0.781250, val_f1: 0.7708, test_f1: 0.8237 - Epoch: 95, Loss: 0.1979, train_time: 0.472037, val_time: 0.393535 train_f1: 0.775000, val_f1: 0.7708, test_f1: 0.8236 - Epoch: 96, Loss: 0.1976, train_time: 0.458223, val_time: 0.402284 train_f1: 0.781250, val_f1: 0.7750, test_f1: 0.8331 - Epoch: 97, Loss: 0.1968, train_time: 0.470268, val_time: 0.398442 train_f1: 0.800000, val_f1: 0.7958, test_f1: 0.8361 - Epoch: 98, Loss: 0.1975, train_time: 0.445167, val_time: 0.391435 train_f1: 0.806250, val_f1: 0.8000, test_f1: 0.8359 - Epoch: 99, Loss: 0.1983, train_time: 0.458230, val_time: 0.385353 train_f1: 0.806250, val_f1: 0.7958, test_f1: 0.8346 - Epoch: 100, Loss: 0.1988, train_time: 0.450563, val_time: 0.386776 train_f1: 0.806250, val_f1: 0.7917, test_f1: 0.8421 - Epoch: 101, Loss: 0.2096, train_time: 0.455066, val_time: 0.392220 train_f1: 0.831250, val_f1: 0.8042, test_f1: 0.8541 - Epoch: 102, Loss: 0.2018, train_time: 0.443387, val_time: 0.394153 train_f1: 0.837500, val_f1: 0.8042, test_f1: 0.8643 - Epoch: 103, Loss: 0.2016, train_time: 0.439255, val_time: 0.387634 train_f1: 0.843750, val_f1: 0.8208, test_f1: 0.8666 - Epoch: 104, Loss: 0.1948, train_time: 0.451777, val_time: 0.386899 train_f1: 0.856250, val_f1: 0.8458, test_f1: 0.8719 - Epoch: 105, Loss: 0.1925, train_time: 0.449302, val_time: 0.389077 train_f1: 0.862500, val_f1: 0.8500, test_f1: 0.8752 - Epoch: 106, Loss: 0.1956, train_time: 0.456920, val_time: 0.393086 train_f1: 0.881250, val_f1: 0.8542, test_f1: 0.8782 - Epoch: 107, Loss: 0.1946, train_time: 0.445179, val_time: 0.389073 train_f1: 0.887500, val_f1: 0.8583, test_f1: 0.8834 - Epoch: 108, Loss: 0.1937, train_time: 0.449708, val_time: 0.388252 train_f1: 0.862500, val_f1: 0.8375, test_f1: 0.8806 - Epoch: 109, Loss: 0.1966, train_time: 0.453591, val_time: 0.392260 train_f1: 0.875000, val_f1: 0.8333, test_f1: 0.8818 - Epoch: 110, Loss: 0.1943, train_time: 0.447050, val_time: 0.393864 train_f1: 0.862500, val_f1: 0.8292, test_f1: 0.8812 - Epoch: 111, Loss: 0.1980, train_time: 0.450165, val_time: 0.394744 train_f1: 0.868750, val_f1: 0.8458, test_f1: 0.8815 - Epoch: 112, Loss: 0.1938, train_time: 0.451525, val_time: 0.395107 train_f1: 0.887500, val_f1: 0.8625, test_f1: 0.8861 - Epoch: 113, Loss: 0.1863, train_time: 0.452203, val_time: 0.397329 train_f1: 0.887500, val_f1: 0.8667, test_f1: 0.8899 - Epoch: 114, Loss: 0.1842, train_time: 0.447603, val_time: 0.393829 train_f1: 0.881250, val_f1: 0.8708, test_f1: 0.8921 - Epoch: 115, Loss: 0.1914, train_time: 0.445139, val_time: 0.392588 train_f1: 0.881250, val_f1: 0.8708, test_f1: 0.8945 - Epoch: 116, Loss: 0.1858, train_time: 0.449721, val_time: 0.394229 train_f1: 0.881250, val_f1: 0.8792, test_f1: 0.8997 - Epoch: 117, Loss: 0.1931, train_time: 0.457180, val_time: 0.387968 train_f1: 0.881250, val_f1: 0.8792, test_f1: 0.9025 - Epoch: 118, Loss: 0.1994, train_time: 0.448175, val_time: 0.396640 train_f1: 0.887500, val_f1: 0.8750, test_f1: 0.9043 - Epoch: 119, Loss: 0.1914, train_time: 0.454013, val_time: 0.405102 train_f1: 0.900000, val_f1: 0.8917, test_f1: 0.9058 - Epoch: 120, Loss: 0.1842, train_time: 0.453593, val_time: 0.393414 train_f1: 0.893750, val_f1: 0.9042, test_f1: 0.9101 - Epoch: 121, Loss: 0.1943, train_time: 0.444421, val_time: 0.392165 train_f1: 0.887500, val_f1: 0.9042, test_f1: 0.9106 - Epoch: 122, Loss: 0.1925, train_time: 0.454191, val_time: 0.390331 train_f1: 0.900000, val_f1: 0.8958, test_f1: 0.9072 - Epoch: 123, Loss: 0.1995, train_time: 0.454651, val_time: 0.390887 train_f1: 0.900000, val_f1: 0.9000, test_f1: 0.9052 - Epoch: 124, Loss: 0.1862, train_time: 0.458681, val_time: 0.404071 train_f1: 0.906250, val_f1: 0.9083, test_f1: 0.9079 - Epoch: 125, Loss: 0.1874, train_time: 0.450586, val_time: 0.394936 train_f1: 0.893750, val_f1: 0.8958, test_f1: 0.9090 - Epoch: 126, Loss: 0.1870, train_time: 0.453429, val_time: 0.391595 train_f1: 0.881250, val_f1: 0.9000, test_f1: 0.9086 - Epoch: 127, Loss: 0.1860, train_time: 0.448238, val_time: 0.393557 train_f1: 0.881250, val_f1: 0.8875, test_f1: 0.9063 - Epoch: 128, Loss: 0.1895, train_time: 0.456745, val_time: 0.393320 train_f1: 0.887500, val_f1: 0.8917, test_f1: 0.9077 - Epoch: 129, Loss: 0.1901, train_time: 0.442648, val_time: 0.390761 train_f1: 0.893750, val_f1: 0.8958, test_f1: 0.9062 - Epoch: 130, Loss: 0.1825, train_time: 0.458853, val_time: 0.408519 train_f1: 0.900000, val_f1: 0.9083, test_f1: 0.9094 - Epoch: 131, Loss: 0.1732, train_time: 0.448754, val_time: 0.391863 train_f1: 0.906250, val_f1: 0.9042, test_f1: 0.9123 - Epoch: 132, Loss: 0.1930, train_time: 0.453109, val_time: 0.391353 train_f1: 0.900000, val_f1: 0.9125, test_f1: 0.9109 - Epoch: 133, Loss: 0.1775, train_time: 0.441247, val_time: 0.392522 train_f1: 0.900000, val_f1: 0.9083, test_f1: 0.9099 - Epoch: 134, Loss: 0.1748, train_time: 0.443449, val_time: 0.385535 train_f1: 0.887500, val_f1: 0.9083, test_f1: 0.9098 - Epoch: 135, Loss: 0.1792, train_time: 0.442838, val_time: 0.385317 train_f1: 0.893750, val_f1: 0.9042, test_f1: 0.9084 - Epoch: 136, Loss: 0.1859, train_time: 0.442256, val_time: 0.385644 train_f1: 0.881250, val_f1: 0.9000, test_f1: 0.9055 - Epoch: 137, Loss: 0.1817, train_time: 0.460867, val_time: 0.385437 train_f1: 0.893750, val_f1: 0.9042, test_f1: 0.9077 - Epoch: 138, Loss: 0.1713, train_time: 0.445950, val_time: 0.384954 train_f1: 0.887500, val_f1: 0.9125, test_f1: 0.9139 - Epoch: 139, Loss: 0.1827, train_time: 0.440833, val_time: 0.388276 train_f1: 0.900000, val_f1: 0.9167, test_f1: 0.9152 - Epoch: 140, Loss: 0.1790, train_time: 0.453825, val_time: 0.390597 train_f1: 0.906250, val_f1: 0.9208, test_f1: 0.9139 - Epoch: 141, Loss: 0.1812, train_time: 0.457757, val_time: 0.384416 train_f1: 0.906250, val_f1: 0.9208, test_f1: 0.9124 - Epoch: 142, Loss: 0.1765, train_time: 0.441343, val_time: 0.397051 train_f1: 0.900000, val_f1: 0.9208, test_f1: 0.9117 - Epoch: 143, Loss: 0.1782, train_time: 0.441256, val_time: 0.388266 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9138 - Epoch: 144, Loss: 0.1815, train_time: 0.443241, val_time: 0.390542 train_f1: 0.906250, val_f1: 0.9167, test_f1: 0.9114 - Epoch: 145, Loss: 0.1761, train_time: 0.440742, val_time: 0.393586 train_f1: 0.893750, val_f1: 0.9167, test_f1: 0.9112 - Epoch: 146, Loss: 0.1690, train_time: 0.463384, val_time: 0.391230 train_f1: 0.887500, val_f1: 0.9125, test_f1: 0.9117 - Epoch: 147, Loss: 0.1746, train_time: 0.444694, val_time: 0.383637 train_f1: 0.887500, val_f1: 0.9083, test_f1: 0.9138 - Epoch: 148, Loss: 0.1698, train_time: 0.447705, val_time: 0.386101 train_f1: 0.906250, val_f1: 0.9250, test_f1: 0.9113 - Epoch: 149, Loss: 0.1705, train_time: 0.458272, val_time: 0.383065 train_f1: 0.906250, val_f1: 0.9250, test_f1: 0.9120 - Epoch: 150, Loss: 0.1747, train_time: 0.452806, val_time: 0.383158 train_f1: 0.912500, val_f1: 0.9083, test_f1: 0.9119 - Epoch: 151, Loss: 0.1856, train_time: 0.442148, val_time: 0.391732 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9127 - Epoch: 152, Loss: 0.1704, train_time: 0.460733, val_time: 0.391192 train_f1: 0.900000, val_f1: 0.9250, test_f1: 0.9128 - Epoch: 153, Loss: 0.1810, train_time: 0.463810, val_time: 0.395869 train_f1: 0.906250, val_f1: 0.9125, test_f1: 0.9163 - Epoch: 154, Loss: 0.1779, train_time: 0.452375, val_time: 0.392212 train_f1: 0.900000, val_f1: 0.9042, test_f1: 0.9150 - Epoch: 155, Loss: 0.1780, train_time: 0.449106, val_time: 0.397114 train_f1: 0.900000, val_f1: 0.9167, test_f1: 0.9094 - Epoch: 156, Loss: 0.1748, train_time: 0.456670, val_time: 0.393993 train_f1: 0.887500, val_f1: 0.9083, test_f1: 0.9011 - Epoch: 157, Loss: 0.1814, train_time: 0.458913, val_time: 0.390159 train_f1: 0.900000, val_f1: 0.9208, test_f1: 0.9074 - Epoch: 158, Loss: 0.1699, train_time: 0.459164, val_time: 0.391982 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9146 - Epoch: 159, Loss: 0.1733, train_time: 0.464545, val_time: 0.391483 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9168 - Epoch: 160, Loss: 0.1764, train_time: 0.446652, val_time: 0.384727 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9119 - Epoch: 161, Loss: 0.1807, train_time: 0.461709, val_time: 0.404569 train_f1: 0.900000, val_f1: 0.9250, test_f1: 0.9097 - Epoch: 162, Loss: 0.1779, train_time: 0.457925, val_time: 0.394702 train_f1: 0.912500, val_f1: 0.9250, test_f1: 0.9153 - Epoch: 163, Loss: 0.1716, train_time: 0.461708, val_time: 0.395214 train_f1: 0.906250, val_f1: 0.9250, test_f1: 0.9164 - Epoch: 164, Loss: 0.1807, train_time: 0.451459, val_time: 0.390888 train_f1: 0.893750, val_f1: 0.9208, test_f1: 0.9103 - Epoch: 165, Loss: 0.1651, train_time: 0.453680, val_time: 0.393294 train_f1: 0.887500, val_f1: 0.9042, test_f1: 0.8989 - Epoch: 166, Loss: 0.1736, train_time: 0.454021, val_time: 0.388737 train_f1: 0.900000, val_f1: 0.9125, test_f1: 0.9025 - Epoch: 167, Loss: 0.1806, train_time: 0.448348, val_time: 0.386748 train_f1: 0.900000, val_f1: 0.9208, test_f1: 0.9149 - Epoch: 168, Loss: 0.1704, train_time: 0.441684, val_time: 0.385814 train_f1: 0.918750, val_f1: 0.9167, test_f1: 0.9157 - Epoch: 169, Loss: 0.1709, train_time: 0.474505, val_time: 0.394620 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9114 - Epoch: 170, Loss: 0.1702, train_time: 0.457869, val_time: 0.399448 train_f1: 0.912500, val_f1: 0.9167, test_f1: 0.9054 - Epoch: 171, Loss: 0.1792, train_time: 0.469363, val_time: 0.391629 train_f1: 0.906250, val_f1: 0.9167, test_f1: 0.9131 - Epoch: 172, Loss: 0.1697, train_time: 0.466157, val_time: 0.391788 train_f1: 0.900000, val_f1: 0.9292, test_f1: 0.9167 - Epoch: 173, Loss: 0.1699, train_time: 0.447252, val_time: 0.400941 train_f1: 0.900000, val_f1: 0.9208, test_f1: 0.9172 - Epoch: 174, Loss: 0.1690, train_time: 0.456203, val_time: 0.390292 train_f1: 0.906250, val_f1: 0.9208, test_f1: 0.9141 - Epoch: 175, Loss: 0.1658, train_time: 0.463908, val_time: 0.394521 train_f1: 0.906250, val_f1: 0.9208, test_f1: 0.9117 - Epoch: 176, Loss: 0.1634, train_time: 0.461511, val_time: 0.399145 train_f1: 0.906250, val_f1: 0.9250, test_f1: 0.9130 - Epoch: 177, Loss: 0.1730, train_time: 0.467495, val_time: 0.399415 train_f1: 0.918750, val_f1: 0.9333, test_f1: 0.9156 - Epoch: 178, Loss: 0.1731, train_time: 0.452674, val_time: 0.387287 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9148 - Epoch: 179, Loss: 0.1635, train_time: 0.443344, val_time: 0.389657 train_f1: 0.912500, val_f1: 0.9250, test_f1: 0.9163 - Epoch: 180, Loss: 0.1631, train_time: 0.458120, val_time: 0.392209 train_f1: 0.912500, val_f1: 0.9250, test_f1: 0.9160 - Epoch: 181, Loss: 0.1677, train_time: 0.458378, val_time: 0.406659 train_f1: 0.925000, val_f1: 0.9250, test_f1: 0.9181 - Epoch: 182, Loss: 0.1754, train_time: 0.471653, val_time: 0.396353 train_f1: 0.918750, val_f1: 0.9333, test_f1: 0.9178 - Epoch: 183, Loss: 0.1710, train_time: 0.471277, val_time: 0.400120 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9127 - Epoch: 184, Loss: 0.1673, train_time: 0.460885, val_time: 0.393499 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9135 - Epoch: 185, Loss: 0.1688, train_time: 0.461851, val_time: 0.394523 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9152 - Epoch: 186, Loss: 0.1677, train_time: 0.454201, val_time: 0.398099 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9175 - Epoch: 187, Loss: 0.1689, train_time: 0.464801, val_time: 0.388531 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9181 - Epoch: 188, Loss: 0.1758, train_time: 0.450999, val_time: 0.400836 train_f1: 0.912500, val_f1: 0.9292, test_f1: 0.9128 - Epoch: 189, Loss: 0.1644, train_time: 0.458503, val_time: 0.401884 train_f1: 0.912500, val_f1: 0.9250, test_f1: 0.9130 - Epoch: 190, Loss: 0.1640, train_time: 0.461937, val_time: 0.391222 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9146 - Epoch: 191, Loss: 0.1650, train_time: 0.447526, val_time: 0.402985 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9160 - Epoch: 192, Loss: 0.1739, train_time: 0.453922, val_time: 0.397233 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9167 - Epoch: 193, Loss: 0.1608, train_time: 0.455353, val_time: 0.389747 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9172 - Epoch: 194, Loss: 0.1708, train_time: 0.451057, val_time: 0.390553 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9186 - Epoch: 195, Loss: 0.1651, train_time: 0.461481, val_time: 0.405649 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9178 - Epoch: 196, Loss: 0.1746, train_time: 0.474096, val_time: 0.398049 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9142 - Epoch: 197, Loss: 0.1765, train_time: 0.463416, val_time: 0.396694 train_f1: 0.918750, val_f1: 0.9208, test_f1: 0.9109 - Epoch: 198, Loss: 0.1627, train_time: 0.466215, val_time: 0.409122 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9143 - Epoch: 199, Loss: 0.1808, train_time: 0.461198, val_time: 0.399740 train_f1: 0.925000, val_f1: 0.9250, test_f1: 0.9177 - Epoch: 200, Loss: 0.1640, train_time: 0.447127, val_time: 0.395700 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9175 - Epoch: 201, Loss: 0.1711, train_time: 0.462343, val_time: 0.392432 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9179 - Epoch: 202, Loss: 0.1652, train_time: 0.448992, val_time: 0.395874 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9156 - Epoch: 203, Loss: 0.1673, train_time: 0.457323, val_time: 0.402517 train_f1: 0.918750, val_f1: 0.9333, test_f1: 0.9114 - Epoch: 204, Loss: 0.1728, train_time: 0.475780, val_time: 0.397789 train_f1: 0.912500, val_f1: 0.9292, test_f1: 0.9108 - Epoch: 205, Loss: 0.1666, train_time: 0.577320, val_time: 0.396043 train_f1: 0.925000, val_f1: 0.9250, test_f1: 0.9153 - Epoch: 206, Loss: 0.1694, train_time: 0.448983, val_time: 0.394768 train_f1: 0.925000, val_f1: 0.9167, test_f1: 0.9159 - Epoch: 207, Loss: 0.1700, train_time: 0.457281, val_time: 0.398448 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9161 - Epoch: 208, Loss: 0.1704, train_time: 0.460470, val_time: 0.397538 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9087 - Epoch: 209, Loss: 0.1659, train_time: 0.445487, val_time: 0.409231 train_f1: 0.918750, val_f1: 0.9250, test_f1: 0.9121 - Epoch: 210, Loss: 0.1623, train_time: 0.458501, val_time: 0.438757 train_f1: 0.918750, val_f1: 0.9208, test_f1: 0.9167 - Epoch: 211, Loss: 0.1649, train_time: 0.468280, val_time: 0.401017 train_f1: 0.918750, val_f1: 0.9208, test_f1: 0.9164 - Epoch: 212, Loss: 0.1664, train_time: 0.463500, val_time: 0.397718 train_f1: 0.918750, val_f1: 0.9292, test_f1: 0.9123 - Epoch: 213, Loss: 0.1639, train_time: 0.465236, val_time: 0.397989 train_f1: 0.918750, val_f1: 0.9333, test_f1: 0.9123 - Epoch: 214, Loss: 0.1687, train_time: 0.449840, val_time: 0.397202 train_f1: 0.918750, val_f1: 0.9333, test_f1: 0.9181 - Epoch: 215, Loss: 0.1653, train_time: 0.452932, val_time: 0.396798 train_f1: 0.943750, val_f1: 0.9167, test_f1: 0.9150 - Epoch: 216, Loss: 0.1697, train_time: 0.458518, val_time: 0.392985 train_f1: 0.925000, val_f1: 0.9208, test_f1: 0.9135 - Epoch: 217, Loss: 0.1606, train_time: 0.461026, val_time: 0.395124 train_f1: 0.918750, val_f1: 0.9167, test_f1: 0.9073 - Epoch: 218, Loss: 0.1631, train_time: 0.467454, val_time: 0.393200 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9120 - Epoch: 219, Loss: 0.1690, train_time: 0.466931, val_time: 0.401497 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9188 - Epoch: 220, Loss: 0.1655, train_time: 0.465267, val_time: 0.390967 train_f1: 0.931250, val_f1: 0.9250, test_f1: 0.9159 - Epoch: 221, Loss: 0.1717, train_time: 0.461511, val_time: 0.402631 train_f1: 0.918750, val_f1: 0.9208, test_f1: 0.9132 - Epoch: 222, Loss: 0.1703, train_time: 0.463001, val_time: 0.410113 train_f1: 0.925000, val_f1: 0.9208, test_f1: 0.9121 - Epoch: 223, Loss: 0.1641, train_time: 0.464444, val_time: 0.408762 train_f1: 0.925000, val_f1: 0.9167, test_f1: 0.9116 - Epoch: 224, Loss: 0.1609, train_time: 0.480095, val_time: 0.390661 train_f1: 0.925000, val_f1: 0.9250, test_f1: 0.9148 - Epoch: 225, Loss: 0.1594, train_time: 0.471180, val_time: 0.403784 train_f1: 0.912500, val_f1: 0.9208, test_f1: 0.9130 - Epoch: 226, Loss: 0.1701, train_time: 0.457036, val_time: 0.391104 train_f1: 0.925000, val_f1: 0.9208, test_f1: 0.9159 - Epoch: 227, Loss: 0.1686, train_time: 0.457494, val_time: 0.394581 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9199 - Epoch: 228, Loss: 0.1579, train_time: 0.455626, val_time: 0.399623 train_f1: 0.925000, val_f1: 0.9167, test_f1: 0.9172 - Epoch: 229, Loss: 0.1706, train_time: 0.450535, val_time: 0.388104 train_f1: 0.918750, val_f1: 0.9375, test_f1: 0.9130 - Epoch: 230, Loss: 0.1676, train_time: 0.463952, val_time: 0.401608 train_f1: 0.912500, val_f1: 0.9250, test_f1: 0.9117 - Epoch: 231, Loss: 0.1706, train_time: 0.464486, val_time: 0.399394 train_f1: 0.925000, val_f1: 0.9208, test_f1: 0.9166 - Epoch: 232, Loss: 0.1583, train_time: 0.469163, val_time: 0.407946 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9186 - Epoch: 233, Loss: 0.1569, train_time: 0.477242, val_time: 0.397354 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9167 - Epoch: 234, Loss: 0.1691, train_time: 0.472784, val_time: 0.401857 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9119 - Epoch: 235, Loss: 0.1595, train_time: 0.479847, val_time: 0.411346 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9156 - Epoch: 236, Loss: 0.1631, train_time: 0.489919, val_time: 0.397272 train_f1: 0.925000, val_f1: 0.9208, test_f1: 0.9171 - Epoch: 237, Loss: 0.1681, train_time: 0.465526, val_time: 0.405837 train_f1: 0.925000, val_f1: 0.9125, test_f1: 0.9159 - Epoch: 238, Loss: 0.1655, train_time: 0.457426, val_time: 0.401923 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9166 - Epoch: 239, Loss: 0.1615, train_time: 0.460302, val_time: 0.412083 train_f1: 0.925000, val_f1: 0.9375, test_f1: 0.9141 - Epoch: 240, Loss: 0.1624, train_time: 0.455020, val_time: 0.402286 train_f1: 0.925000, val_f1: 0.9167, test_f1: 0.9174 - Epoch: 241, Loss: 0.1631, train_time: 0.460418, val_time: 0.406516 train_f1: 0.931250, val_f1: 0.9250, test_f1: 0.9094 - Epoch: 242, Loss: 0.1582, train_time: 0.470588, val_time: 0.397048 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9137 - Epoch: 243, Loss: 0.1585, train_time: 0.454023, val_time: 0.401365 train_f1: 0.937500, val_f1: 0.9333, test_f1: 0.9203 - Epoch: 244, Loss: 0.1588, train_time: 0.457538, val_time: 0.421644 train_f1: 0.925000, val_f1: 0.9375, test_f1: 0.9203 - Epoch: 245, Loss: 0.1707, train_time: 0.455922, val_time: 0.404543 train_f1: 0.925000, val_f1: 0.9417, test_f1: 0.9153 - Epoch: 246, Loss: 0.1666, train_time: 0.461825, val_time: 0.403364 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9123 - Epoch: 247, Loss: 0.1643, train_time: 0.452025, val_time: 0.400922 train_f1: 0.931250, val_f1: 0.9292, test_f1: 0.9159 - Epoch: 248, Loss: 0.1585, train_time: 0.455410, val_time: 0.401873 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9179 - Epoch: 249, Loss: 0.1560, train_time: 0.467158, val_time: 0.397890 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9168 - Epoch: 250, Loss: 0.1592, train_time: 0.456920, val_time: 0.396781 train_f1: 0.937500, val_f1: 0.9333, test_f1: 0.9143 - Epoch: 251, Loss: 0.1583, train_time: 0.462412, val_time: 0.400638 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9110 - Epoch: 252, Loss: 0.1582, train_time: 0.454033, val_time: 0.415506 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9120 - Epoch: 253, Loss: 0.1674, train_time: 0.450033, val_time: 0.406660 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9186 - Epoch: 254, Loss: 0.1605, train_time: 0.457658, val_time: 0.418767 train_f1: 0.937500, val_f1: 0.9250, test_f1: 0.9193 - Epoch: 255, Loss: 0.1581, train_time: 0.459408, val_time: 0.395032 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9182 - Epoch: 256, Loss: 0.1552, train_time: 0.461952, val_time: 0.395385 train_f1: 0.931250, val_f1: 0.9250, test_f1: 0.9160 - Epoch: 257, Loss: 0.1520, train_time: 0.466273, val_time: 0.402267 train_f1: 0.931250, val_f1: 0.9250, test_f1: 0.9149 - Epoch: 258, Loss: 0.1639, train_time: 0.472842, val_time: 0.402324 train_f1: 0.925000, val_f1: 0.9292, test_f1: 0.9143 - Epoch: 259, Loss: 0.1655, train_time: 0.467757, val_time: 0.405136 train_f1: 0.937500, val_f1: 0.9250, test_f1: 0.9154 - Epoch: 260, Loss: 0.1515, train_time: 0.452410, val_time: 0.391822 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9167 - Epoch: 261, Loss: 0.1589, train_time: 0.461282, val_time: 0.389266 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9167 - Epoch: 262, Loss: 0.1640, train_time: 0.451252, val_time: 0.392045 train_f1: 0.925000, val_f1: 0.9250, test_f1: 0.9116 - Epoch: 263, Loss: 0.1592, train_time: 0.463023, val_time: 0.391295 train_f1: 0.918750, val_f1: 0.9208, test_f1: 0.9117 - Epoch: 264, Loss: 0.1638, train_time: 0.463214, val_time: 0.395226 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9186 - Epoch: 265, Loss: 0.1581, train_time: 0.471109, val_time: 0.393808 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9175 - Epoch: 266, Loss: 0.1588, train_time: 0.467802, val_time: 0.396646 train_f1: 0.931250, val_f1: 0.9292, test_f1: 0.9139 - Epoch: 267, Loss: 0.1590, train_time: 0.459547, val_time: 0.391811 train_f1: 0.937500, val_f1: 0.9208, test_f1: 0.9108 - Epoch: 268, Loss: 0.1716, train_time: 0.459391, val_time: 0.396645 train_f1: 0.931250, val_f1: 0.9208, test_f1: 0.9130 - Epoch: 269, Loss: 0.1615, train_time: 0.468447, val_time: 0.389689 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9163 - Epoch: 270, Loss: 0.1581, train_time: 0.472106, val_time: 0.399589 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9193 - Epoch: 271, Loss: 0.1576, train_time: 0.471457, val_time: 0.399265 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9183 - Epoch: 272, Loss: 0.1539, train_time: 0.454661, val_time: 0.392248 train_f1: 0.937500, val_f1: 0.9333, test_f1: 0.9142 - Epoch: 273, Loss: 0.1513, train_time: 0.451976, val_time: 0.389326 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9149 - Epoch: 274, Loss: 0.1585, train_time: 0.451466, val_time: 0.386282 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9181 - Epoch: 275, Loss: 0.1636, train_time: 0.443387, val_time: 0.386783 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9201 - Epoch: 276, Loss: 0.1533, train_time: 0.451974, val_time: 0.389416 train_f1: 0.925000, val_f1: 0.9333, test_f1: 0.9131 - Epoch: 277, Loss: 0.1577, train_time: 0.461919, val_time: 0.387502 train_f1: 0.925000, val_f1: 0.9375, test_f1: 0.9127 - Epoch: 278, Loss: 0.1615, train_time: 0.438472, val_time: 0.393333 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9159 - Epoch: 279, Loss: 0.1615, train_time: 0.444278, val_time: 0.390064 train_f1: 0.937500, val_f1: 0.9333, test_f1: 0.9156 - Epoch: 280, Loss: 0.1543, train_time: 0.458565, val_time: 0.420598 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9135 - Epoch: 281, Loss: 0.1566, train_time: 0.452781, val_time: 0.396424 train_f1: 0.931250, val_f1: 0.9292, test_f1: 0.9099 - Epoch: 282, Loss: 0.1554, train_time: 0.448304, val_time: 0.395016 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9124 - Epoch: 283, Loss: 0.1640, train_time: 0.458192, val_time: 0.390087 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9156 - Epoch: 284, Loss: 0.1553, train_time: 0.440214, val_time: 0.383308 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9167 - Epoch: 285, Loss: 0.1521, train_time: 0.448102, val_time: 0.397962 train_f1: 0.931250, val_f1: 0.9417, test_f1: 0.9164 - Epoch: 286, Loss: 0.1460, train_time: 0.453427, val_time: 0.399234 train_f1: 0.931250, val_f1: 0.9417, test_f1: 0.9132 - Epoch: 287, Loss: 0.1506, train_time: 0.446965, val_time: 0.386775 train_f1: 0.931250, val_f1: 0.9208, test_f1: 0.9138 - Epoch: 288, Loss: 0.1559, train_time: 0.443628, val_time: 0.397555 train_f1: 0.937500, val_f1: 0.9292, test_f1: 0.9139 - Epoch: 289, Loss: 0.1589, train_time: 0.466519, val_time: 0.389501 train_f1: 0.931250, val_f1: 0.9333, test_f1: 0.9149 - Epoch: 290, Loss: 0.1534, train_time: 0.441172, val_time: 0.388616 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9174 - Epoch: 291, Loss: 0.1547, train_time: 0.447603, val_time: 0.392236 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9171 - Epoch: 292, Loss: 0.1516, train_time: 0.445864, val_time: 0.393170 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9168 - Epoch: 293, Loss: 0.1585, train_time: 0.438612, val_time: 0.390646 train_f1: 0.943750, val_f1: 0.9292, test_f1: 0.9138 - Epoch: 294, Loss: 0.1536, train_time: 0.445481, val_time: 0.400595 train_f1: 0.943750, val_f1: 0.9333, test_f1: 0.9120 - Epoch: 295, Loss: 0.1552, train_time: 0.468782, val_time: 0.403011 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9142 - Epoch: 296, Loss: 0.1539, train_time: 0.456322, val_time: 0.387627 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9183 - Epoch: 297, Loss: 0.1518, train_time: 0.464343, val_time: 0.404478 train_f1: 0.931250, val_f1: 0.9375, test_f1: 0.9175 - Epoch: 298, Loss: 0.1562, train_time: 0.451021, val_time: 0.390593 train_f1: 0.937500, val_f1: 0.9333, test_f1: 0.9097 - Epoch: 299, Loss: 0.1632, train_time: 0.462334, val_time: 0.386922 train_f1: 0.937500, val_f1: 0.9375, test_f1: 0.9182 - Epoch: 300, Loss: 0.1505, train_time: 0.447994, val_time: 0.404172 train_f1: 0.943750, val_f1: 0.9250, test_f1: 0.9174 -``` + * NASA 经调参，300 epoch, Amazon Photo, F1 ~ 0.92 左右。 diff --git a/agl/python/examples/nasa/nasa.py b/agl/python/examples/nasa/nasa.py index b28432a..b3e7052 100644 --- a/agl/python/examples/nasa/nasa.py +++ b/agl/python/examples/nasa/nasa.py @@ -15,6 +15,17 @@ from agl.python.model.utils.nasa_utils import * + +def setup_seed(seed=2023): + torch.manual_seed(seed) + torch.cuda.manual_seed_all(seed) + np.random.seed(seed) + random.seed(seed) + torch.backends.cudnn.deterministic = True + + +setup_seed() + parser = argparse.ArgumentParser() parser.add_argument("--lr", type=float, default=0.01, help="Initial learning rate.") parser.add_argument("--weight_decay", type=float, default=1e-3, help="Weight decay") @@ -25,6 +36,7 @@ parser.add_argument("--in_channels", type=int, default=745, help="in channels of GNN") parser.add_argument("--out_channels", type=int, default=8, help="out channels of GNN") parser.add_argument("--num_layers", type=int, default=2, help="layer number of GNN") +parser.add_argument("--max_epoch", type=int, default=300, help="max training epoch") args = parser.parse_args() # step 1: 构建dataset @@ -164,7 +176,7 @@ def test(loader, flag="test"): print("Training!") -for epoch in range(1, 500): +for epoch in range(args.max_epoch): t0 = time.time() loss = train(train_loader) diff --git a/agl/python/examples/pagnn/ReadMe.md b/agl/python/examples/pagnn/ReadMe.md index bba61e3..6c2901f 100644 --- a/agl/python/examples/pagnn/ReadMe.md +++ b/agl/python/examples/pagnn/ReadMe.md @@ -3,14 +3,16 @@ > Paper: Inductive Link Prediction with Interactive Structure Learning on Attributed Graph https://2021.ecmlpkdd.org/wp-content/uploads/2021/07/sub_635.pdf -由于link类算法的样本数量过多，只能在分布式模式（比如yarn）运行，为了方便不想搭建yarn集群的用户，我们提供了采样好的图样本下载地址为https://drive.google.com/drive/folders/11VphpI6rgvKzfElF9slp2D9qSJvT-big?usp=sharing，用户可直接进行模型训练 ### 数据下载： - 从Facebook: https://docs.google.com/uc?export=download&id=12aJWAGCM4IvdGI2fiydDNyWzViEOLZH8&confirm=t下载数据,放到data_process/facebook/raw目录下 +从Facebook: https://docs.google.com/uc?export=download&id=12aJWAGCM4IvdGI2fiydDNyWzViEOLZH8&confirm=t 下载数据， +放到data_process/facebook/raw目录下，切换到data_process目录下执行python data_prepare.py生成点表、边表等 ### 数据预处理与子图采样： - 运行submit.sh进行数据预处理和spark采样,得到训练集测试集验证集 -### 模型 - python pagnn_facebook.py +运行submit.sh进行数据预处理和spark采样,得到训练集测试集验证集 +由于link类算法的样本数量过多，只能在分布式模式（比如yarn）运行，为了方便不想搭建yarn集群的用户， +我们提供了采样好的图样本下载地址为 https://drive.google.com/drive/folders/11VphpI6rgvKzfElF9slp2D9qSJvT-big?usp=sharing ，用户可直接进行模型训练 +### 模型训练 +python pagnn_facebook.py ### Model: @@ -19,7 +21,7 @@ the information of source nodes to the subgraph, and target nodes will perceive the structural information of the subgraph during aggregation(i.e. paths, common neighbors). # Benchmark - +``` Facebook AUC~99.0% Train 75%, Val 5%, Test 20%, embedding_size = 32, n_hops=2 Logs: @@ -73,57 +75,4 @@ Epoch: 24, Loss: 0.0753, Val_AUC: 0.9903, Test_AUC: 0.9905, Final_AUC: 0.9905, t 103.919959 Epoch: 25, Loss: 0.0748, Val_AUC: 0.9903, Test_AUC: 0.9906, Final_AUC: 0.9905, train_time: 331.601013, val_time: 103.640856 - -Pubmed AUC~94.1% -Train 75%, Val 5%, Test 20%, embedding_size = 32, n_hops=2 -Logs: -Epoch: 01, Loss: 0.4376, Val_AUC: 0.8985, Test_AUC: 0.9038, Final_AUC: 0.9038, train_time: 325.235858, val_time: -84.538931 -Epoch: 02, Loss: 0.3512, Val_AUC: 0.9222, Test_AUC: 0.9291, Final_AUC: 0.9291, train_time: 322.038818, val_time: -84.642969 -Epoch: 03, Loss: 0.3311, Val_AUC: 0.9246, Test_AUC: 0.9314, Final_AUC: 0.9314, train_time: 323.316217, val_time: -84.442774 -Epoch: 04, Loss: 0.3263, Val_AUC: 0.9153, Test_AUC: 0.9223, Final_AUC: 0.9314, train_time: 324.821595, val_time: -84.640268 -Epoch: 05, Loss: 0.3343, Val_AUC: 0.9252, Test_AUC: 0.9321, Final_AUC: 0.9321, train_time: 323.989801, val_time: -84.509204 -Epoch: 06, Loss: 0.3374, Val_AUC: 0.8716, Test_AUC: 0.8790, Final_AUC: 0.9321, train_time: 322.228783, val_time: -84.002366 -Epoch: 07, Loss: 0.4274, Val_AUC: 0.8930, Test_AUC: 0.8998, Final_AUC: 0.9321, train_time: 319.993209, val_time: -85.039291 -Epoch: 08, Loss: 0.4441, Val_AUC: 0.8545, Test_AUC: 0.8609, Final_AUC: 0.9321, train_time: 325.361055, val_time: -85.668372 -Epoch: 09, Loss: 0.3778, Val_AUC: 0.9176, Test_AUC: 0.9239, Final_AUC: 0.9321, train_time: 324.677065, val_time: -84.033798 -Epoch: 10, Loss: 0.3366, Val_AUC: 0.9248, Test_AUC: 0.9306, Final_AUC: 0.9321, train_time: 323.139526, val_time: -84.098302 -Epoch: 11, Loss: 0.3240, Val_AUC: 0.9235, Test_AUC: 0.9295, Final_AUC: 0.9321, train_time: 325.894223, val_time: -84.128854 -Epoch: 12, Loss: 0.3202, Val_AUC: 0.9297, Test_AUC: 0.9348, Final_AUC: 0.9348, train_time: 325.761913, val_time: -84.220473 -Epoch: 13, Loss: 0.3115, Val_AUC: 0.9306, Test_AUC: 0.9360, Final_AUC: 0.9360, train_time: 326.451049, val_time: -84.056654 -Epoch: 14, Loss: 0.3060, Val_AUC: 0.9325, Test_AUC: 0.9373, Final_AUC: 0.9373, train_time: 324.388285, val_time: -84.209103 -Epoch: 15, Loss: 0.3040, Val_AUC: 0.9294, Test_AUC: 0.9344, Final_AUC: 0.9373, train_time: 323.140053, val_time: -85.050226 -Epoch: 16, Loss: 0.3008, Val_AUC: 0.9326, Test_AUC: 0.9378, Final_AUC: 0.9378, train_time: 321.854683, val_time: -84.415050 -Epoch: 17, Loss: 0.3008, Val_AUC: 0.9327, Test_AUC: 0.9381, Final_AUC: 0.9381, train_time: 321.498435, val_time: -85.207391 -Epoch: 18, Loss: 0.2982, Val_AUC: 0.9349, Test_AUC: 0.9397, Final_AUC: 0.9397, train_time: 322.666984, val_time: -84.518749 -Epoch: 19, Loss: 0.2922, Val_AUC: 0.9363, Test_AUC: 0.9411, Final_AUC: 0.9411, train_time: 321.937428, val_time: -84.985604 -Epoch: 20, Loss: 0.2913, Val_AUC: 0.9344, Test_AUC: 0.9400, Final_AUC: 0.9411, train_time: 321.992191, val_time: -84.927943 -Epoch: 21, Loss: 0.2935, Val_AUC: 0.9261, Test_AUC: 0.9313, Final_AUC: 0.9411, train_time: 323.596506, val_time: -84.102577 -Epoch: 22, Loss: 0.3043, Val_AUC: 0.9315, Test_AUC: 0.9367, Final_AUC: 0.9411, train_time: 321.777333, val_time: -84.588558 -Epoch: 23, Loss: 0.3044, Val_AUC: 0.9314, Test_AUC: 0.9355, Final_AUC: 0.9411, train_time: 322.780236, val_time: -85.008025 -Epoch: 24, Loss: 0.3070, Val_AUC: 0.9297, Test_AUC: 0.9340, Final_AUC: 0.9411, train_time: 322.570970, val_time: -83.989334 -Epoch: 25, Loss: 0.2948, Val_AUC: 0.9332, Test_AUC: 0.9366, Final_AUC: 0.9411, train_time: 322.835144, val_time: -84.995032 +``` diff --git a/agl/python/examples/ssr_lastfm/data_process/data_preprcoss.py b/agl/python/examples/ssr_lastfm/data_process/data_preprcoss.py index b9501ab..12f6e68 100755 --- a/agl/python/examples/ssr_lastfm/data_process/data_preprcoss.py +++ b/agl/python/examples/ssr_lastfm/data_process/data_preprcoss.py @@ -3,7 +3,7 @@ def read_lastfm(): edges = [] - with open("./user_artists.dat", "r") as f: + with open("./data/user_artists.dat", "r") as f: f.readline() for line in f: l = line.strip().split() @@ -16,7 +16,7 @@ def read_lastfm(): ) social_relations = [] - with open("./user_friends.dat", "r") as f: + with open("./data/user_friends.dat", "r") as f: f.readline() for line in f: l = line.strip().split() @@ -55,11 +55,11 @@ def negative_sampling(user_nbrs, edges, item_num): ratio = int(0.8 * len(edges)) train_edges, test_edges = edges[:ratio], edges[ratio:] - with open("./agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv", "w") as f: + with open("./data/agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv", "w") as f: print("node_id,node_feature", file=f) for i in range(user_num + item_num): print(i, i, file=f, sep=",") - with open("./agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv", "w") as f: + with open("./data/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv", "w") as f: print("node1_id,node2_id,edge_id", file=f) for u, i in train_edges: i += user_num @@ -71,7 +71,7 @@ def negative_sampling(user_nbrs, edges, item_num): train_neg_edges = negative_sampling(user_nbrs, train_edges, item_num) test_neg_edges = negative_sampling(user_nbrs, test_edges, item_num) - with open("./agl_gzoo_bmdata_ssr_lastfm_open_source_node_label.csv", "w") as f: + with open("./data/agl_gzoo_bmdata_ssr_lastfm_open_source_node_label.csv", "w") as f: print("node1_id,node2_id,seed,label,train_flag", file=f) for u, i in train_edges: i += user_num diff --git a/agl/python/examples/ssr_lastfm/data_process/run_ssr_link.sh b/agl/python/examples/ssr_lastfm/data_process/run_ssr_link.sh index 3aeff5f..d498a86 100644 --- a/agl/python/examples/ssr_lastfm/data_process/run_ssr_link.sh +++ b/agl/python/examples/ssr_lastfm/data_process/run_ssr_link.sh @@ -3,15 +3,15 @@ cd "$base" python ../../run_spark.py \ --jar_resource_path ../../../../java/target/flatv3-1.0-SNAPSHOT.jar \ - --input_edge_table_name agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv \ - --input_label_table_name agl_gzoo_bmdata_ssr_lastfm_open_source_node_label.csv \ - --input_node_table_name agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv \ - --output_table_name_prefix output_graph_feature \ + --input_edge_table_name data/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv \ + --input_label_table_name data/agl_gzoo_bmdata_ssr_lastfm_open_source_node_label.csv \ + --input_node_table_name data/agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv \ + --output_table_name_prefix data/output_graph_feature \ --neighbor_distance 2 \ --train_flag "train_flag" \ --sample_condition 'random_sampler(limit=50, seed=34, replacement=false)' \ --subgraph_spec "{'node_spec':[{'node_name':'default','id_type':'string','features':[{'name':'node_feature','type':'dense','dim':1,'value':'int64'}]}],'edge_spec':[{'edge_name':'default','n1_name':'default','n2_name':'default','id_type':'string','features':[]}]}" \ --algorithm link_level -cp output_graph_feature_1/part* subgraph_ssr_lastfm_train.csv -cp output_graph_feature_0/part* subgraph_ssr_lastfm_test.csv +cp data/output_graph_feature_1/part* subgraph_ssr_lastfm_train.csv +cp data/output_graph_feature_0/part* subgraph_ssr_lastfm_test.csv diff --git a/agl/python/examples/ssr_lastfm/readme.md b/agl/python/examples/ssr_lastfm/readme.md index 5e7c273..5fad825 100755 --- a/agl/python/examples/ssr_lastfm/readme.md +++ b/agl/python/examples/ssr_lastfm/readme.md @@ -16,10 +16,11 @@ ### 数据预处理 以movielens为例子 #### 开源数据预处理 +从 http://files.grouplens.org/datasets/hetrec2011/hetrec2011-lastfm-2k.zip 下载lastfm数据，创建data_process/data目录,将下载数据解压后放在data目录下，使用下述命令预处理成我们需要的格式。 + ``` python data_preprocess.py ``` -从http://files.grouplens.org/datasets/hetrec2011/hetrec2011-lastfm-2k.zip下载lastfm数据，创建data_process/data目录,将下载数据解压后放在data目录下，使用上述命令预处理成我们需要的格式。 #### 第一阶段数据预处理 首先我们要把原始数据压缩成子图(pb string)的形式，使用data_process/run_ssr_link.sh脚本命令 @@ -84,56 +85,4 @@ python ssr2_ego_eval.py ## BenchMark * 效果 - * ssr 未调参，50 epoch, auc ~0.8左右, prec@15 ~ 0.36 左右 (原论文 0.353), recall@15 ~ 0.54左右 (原论文0.536) - ``` - Epoch: 01, Loss: 0.8043, auc: 0.5971, train_time: 12.107816, val_time: 0.483347 - Epoch: 02, Loss: 0.7139, auc: 0.6542, train_time: 11.733946, val_time: 0.328861 - Epoch: 03, Loss: 0.6792, auc: 0.7151, train_time: 11.480709, val_time: 0.303506 - Epoch: 04, Loss: 0.6656, auc: 0.7291, train_time: 11.522599, val_time: 0.294176 - Epoch: 05, Loss: 0.6560, auc: 0.7250, train_time: 11.682429, val_time: 0.445539 - Epoch: 06, Loss: 0.6206, auc: 0.7493, train_time: 11.510933, val_time: 0.445137 - Epoch: 07, Loss: 0.6160, auc: 0.7491, train_time: 11.684788, val_time: 0.297126 - Epoch: 08, Loss: 0.6176, auc: 0.7612, train_time: 11.519130, val_time: 0.306026 - Epoch: 09, Loss: 0.6303, auc: 0.7688, train_time: 11.692025, val_time: 0.327492 - Epoch: 10, Loss: 0.6187, auc: 0.7706, train_time: 11.634877, val_time: 0.337790 - Epoch: 11, Loss: 0.5952, auc: 0.7768, train_time: 11.609031, val_time: 0.310153 - Epoch: 12, Loss: 0.5875, auc: 0.7667, train_time: 11.909676, val_time: 0.292688 - Epoch: 13, Loss: 0.5945, auc: 0.7681, train_time: 11.806919, val_time: 0.295133 - Epoch: 14, Loss: 0.5876, auc: 0.7827, train_time: 11.788707, val_time: 0.338131 - Epoch: 15, Loss: 0.5818, auc: 0.7793, train_time: 11.589313, val_time: 0.307520 - Epoch: 16, Loss: 0.5770, auc: 0.7782, train_time: 11.764631, val_time: 0.302438 - Epoch: 17, Loss: 0.5698, auc: 0.7869, train_time: 11.454369, val_time: 0.413306 - Epoch: 18, Loss: 0.5636, auc: 0.7918, train_time: 11.743885, val_time: 0.409539 - Epoch: 19, Loss: 0.5648, auc: 0.7912, train_time: 11.727757, val_time: 0.298214 - Epoch: 20, Loss: 0.5618, auc: 0.7923, train_time: 11.660682, val_time: 0.327231 - Epoch: 21, Loss: 0.5530, auc: 0.7943, train_time: 11.615249, val_time: 0.286587 - Epoch: 22, Loss: 0.5508, auc: 0.7883, train_time: 11.253371, val_time: 0.331646 - Epoch: 23, Loss: 0.5518, auc: 0.7927, train_time: 11.484581, val_time: 0.331716 - Epoch: 24, Loss: 0.5418, auc: 0.7972, train_time: 11.528899, val_time: 0.308683 - Epoch: 25, Loss: 0.5456, auc: 0.7990, train_time: 11.929682, val_time: 0.310510 - Epoch: 26, Loss: 0.5350, auc: 0.7929, train_time: 11.604513, val_time: 0.297678 - Epoch: 27, Loss: 0.5394, auc: 0.7989, train_time: 11.453251, val_time: 0.292810 - Epoch: 28, Loss: 0.5320, auc: 0.8010, train_time: 11.624455, val_time: 0.309475 - Epoch: 29, Loss: 0.5338, auc: 0.8009, train_time: 11.762576, val_time: 0.294682 - Epoch: 30, Loss: 0.5297, auc: 0.7962, train_time: 11.660346, val_time: 0.292787 - Epoch: 31, Loss: 0.5274, auc: 0.8027, train_time: 11.862746, val_time: 0.291134 - Epoch: 32, Loss: 0.5272, auc: 0.8037, train_time: 11.743582, val_time: 0.398789 - Epoch: 33, Loss: 0.5226, auc: 0.8007, train_time: 11.650809, val_time: 0.296985 - Epoch: 34, Loss: 0.5226, auc: 0.7967, train_time: 11.597193, val_time: 0.321086 - Epoch: 35, Loss: 0.5177, auc: 0.8026, train_time: 11.647320, val_time: 0.304086 - Epoch: 36, Loss: 0.5195, auc: 0.8025, train_time: 11.715076, val_time: 0.287590 - Epoch: 37, Loss: 0.5158, auc: 0.7962, train_time: 11.835550, val_time: 0.282753 - Epoch: 38, Loss: 0.5152, auc: 0.7973, train_time: 11.643948, val_time: 0.397647 - Epoch: 39, Loss: 0.5096, auc: 0.8003, train_time: 11.644680, val_time: 0.329439 - Epoch: 40, Loss: 0.5094, auc: 0.8017, train_time: 11.574507, val_time: 0.300908 - Epoch: 41, Loss: 0.5082, auc: 0.7962, train_time: 11.725015, val_time: 0.329480 - Epoch: 42, Loss: 0.5032, auc: 0.8004, train_time: 11.562064, val_time: 0.303209 - Epoch: 43, Loss: 0.5043, auc: 0.7983, train_time: 11.420388, val_time: 0.310526 - Epoch: 44, Loss: 0.5006, auc: 0.7941, train_time: 11.590370, val_time: 0.290690 - Epoch: 45, Loss: 0.4993, auc: 0.7979, train_time: 11.677804, val_time: 0.282144 - Epoch: 46, Loss: 0.4959, auc: 0.7978, train_time: 11.377506, val_time: 0.420724 - Epoch: 47, Loss: 0.4935, auc: 0.7913, train_time: 12.186495, val_time: 0.289656 - Epoch: 48, Loss: 0.4935, auc: 0.7926, train_time: 11.643528, val_time: 0.398964 - Epoch: 49, Loss: 0.4906, auc: 0.7975, train_time: 11.857935, val_time: 0.353652 - Epoch: 50, Loss: 0.4884, auc: 0.7921, train_time: 11.917154, val_time: 0.295094 - ``` + * ssr 未调参，50 epoch, auc ~0.8左右, prec@15 ~ 0.36 左右 , recall@15 ~ 0.54左右 \ No newline at end of file diff --git a/agl/python/examples/ssr_lastfm/ssr2_ego_eval.py b/agl/python/examples/ssr_lastfm/ssr2_ego_eval.py index 9d11e71..fd4ad7d 100755 --- a/agl/python/examples/ssr_lastfm/ssr2_ego_eval.py +++ b/agl/python/examples/ssr_lastfm/ssr2_ego_eval.py @@ -35,7 +35,7 @@ def load_edge_data(filename): user_embed = torch.load("result/model2_user_embed.pt").cpu().detach().numpy() item_embed = torch.load("result/model2_item_embed.pt").cpu().detach().numpy() test_data = load_data("./data_process/subgraph_ssr_lastfm_test.csv", filter_label=True) -edge_table = load_edge_data("./data_process/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv") +edge_table = load_edge_data("./data_process/data/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv") print(user_embed.shape, item_embed.shape, test_data.shape, edge_table.shape) edge_dict = {} diff --git a/agl/python/examples/ssr_lastfm/ssr_ego_infer.py b/agl/python/examples/ssr_lastfm/ssr_ego_infer.py index ac7f7ca..b425a17 100755 --- a/agl/python/examples/ssr_lastfm/ssr_ego_infer.py +++ b/agl/python/examples/ssr_lastfm/ssr_ego_infer.py @@ -13,8 +13,8 @@ from ssr_ego import SSRLastfmModel # step 1: 构建dataset -node_file_name = "./data_process/agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv" -edge_file_name = "./data_process/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv" +node_file_name = "./data_process/data/agl_gzoo_bmdata_ssr_lastfm_open_source_node_table.csv" +edge_file_name = "./data_process/data/agl_gzoo_bmdata_ssr_lastfm_open_source_edge_table.csv" # node data set node_data_set = AGLTorchMapBasedDataset( diff --git a/build.sh b/build.sh index 77fa3d7..de4a7f1 100644 --- a/build.sh +++ b/build.sh @@ -1,5 +1,8 @@ #!/usr/bin/env bash +# add safe.directory +git config --global --add safe.directory /graph_ml + set -e PROJECT_DIR=$(cd "$(dirname $0)" && pwd) diff --git a/doc/core/install.md b/doc/core/install.md index 6f76392..27f90f9 100644 --- a/doc/core/install.md +++ b/doc/core/install.md @@ -11,23 +11,23 @@ aglimage/agl:agl-ubuntu-gcc9.4.0-py3.8-cuda11.8-pytorch2.0.1-0825 镜像中包含运行 AGL 所需的依赖，包括Java, Maven, Spark, Pytorch, PyG, Cuda等。 在该镜像的 docker 容器中，用户无需再关注环境配置. -启动 docker 脚本 [参考](../../docker/start_docker_with_image.sh) +## 安装步骤： + +### 1. Clone 代码 ``` -bash start_docker_with_image.sh +git clone https://github.com/TuGraph-family/TuGraph-AntGraphLearning.git ``` -## 编译源码 +### 2. 启动 docker -进入 docker 容器后： - -1. clone code +启动 docker 脚本 [参考](../../docker/start_docker_with_image.sh) ``` -git clone https://github.com/TuGraph-family/TuGraph-AntGraphLearning.git +bash start_docker_with_image.sh ``` -2. Build package +### 3. 编译源码 ``` bash build.sh