-
Notifications
You must be signed in to change notification settings - Fork 62
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Xiang Song
committed
Jun 15, 2024
1 parent
4a1adef
commit 2cee7ff
Showing
3 changed files
with
333 additions
and
8 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
284 changes: 284 additions & 0 deletions
284
tests/end2end-tests/data_process/gen_multi_task_remap_test.py
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,284 @@ | ||
| """ | ||
| Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved. | ||
| Licensed under the Apache License, Version 2.0 (the "License"). | ||
| You may not use this file except in compliance with the License. | ||
| You may obtain a copy of the License at | ||
| http://www.apache.org/licenses/LICENSE-2.0 | ||
| Unless required by applicable law or agreed to in writing, software | ||
| distributed under the License is distributed on an "AS IS" BASIS, | ||
| WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| See the License for the specific language governing permissions and | ||
| limitations under the License. | ||
| """ | ||
| import os | ||
| import argparse | ||
| import json | ||
| import yaml | ||
| from argparse import Namespace | ||
|
|
||
| import numpy as np | ||
| import torch as th | ||
|
|
||
| from graphstorm.config import GSConfig | ||
| from graphstorm.gconstruct.id_map import IdMap | ||
| from graphstorm.model.utils import pad_file_index | ||
|
|
||
| def main(args): | ||
| output_path = args.output | ||
| os.makedirs(output_path, exist_ok=True) | ||
| save_prediction_path = os.path.join(output_path, "predict") | ||
| save_embed_path = os.path.join(output_path, "emb") | ||
| ntype0 = "n0" | ||
| ntype1 = "n1" | ||
| etype0 = ("n0", "access", "n1") | ||
| etype1 = ("n1", "access", "n0") | ||
|
|
||
| # generate dummy graph data | ||
| graph_part_dir = os.path.join(output_path, "graph") | ||
| os.makedirs(graph_part_dir, exist_ok=True) | ||
| part_config = os.path.join(graph_part_dir, "graph.json") | ||
| with open(part_config, 'w', encoding='utf-8') as f: | ||
| json.dump({"graph_name":"dummy"}, f, indent=4) | ||
| os.makedirs(os.path.join(graph_part_dir, "part0"), exist_ok=True) | ||
|
|
||
| task_config = { | ||
| "gsf": { | ||
| "basic": { | ||
| "backend": "gloo", | ||
| "batch_size": 32, | ||
| "part_config": part_config | ||
| }, | ||
| "gnn": { | ||
| "model_encoder_type": "rgcn", | ||
| "num_layers": 1, | ||
| "fanout": "5", | ||
| "hidden_size": 32 | ||
| }, | ||
| "output": { | ||
| "save_prediction_path": save_prediction_path, | ||
| "save_embed_path": save_embed_path | ||
| }, | ||
| "multi_task_learning": [ | ||
| { | ||
| "edge_classification": { | ||
| "target_etype": [ | ||
| ",".join(etype0) | ||
| ], | ||
| "label_field": "test_ec0", | ||
| "num_classes": 1000, | ||
| "batch_size": 64 | ||
| }, | ||
| }, | ||
| { | ||
| "edge_classification": { | ||
| "target_etype": [ | ||
| ",".join(etype1) | ||
| ], | ||
| "label_field": "test_ec1", | ||
| "num_classes": 1000, | ||
| "batch_size": 64 | ||
| }, | ||
| }, | ||
| { | ||
| "node_classification": { | ||
| "target_ntype": ntype0, | ||
| "label_field": "test_nc1", | ||
| "num_classes": 1000, | ||
| "batch_size": 64 | ||
| }, | ||
| }, | ||
| { | ||
| # will be ignored | ||
| "link_prediction": { | ||
| "exclude_training_targets": False | ||
| } | ||
| }, | ||
| { | ||
| # will be ignored | ||
| "reconstruct_node_feat": { | ||
| "target_ntype": ntype0, | ||
| "reconstruct_nfeat_name": "feat" | ||
| } | ||
| } | ||
| ] | ||
| } | ||
| } | ||
| task_config_path = os.path.join(output_path, "task.yaml") | ||
| with open(task_config_path, 'w', encoding='utf-8') as f: | ||
| yaml.dump(task_config, f) | ||
|
|
||
| args = Namespace(yaml_config_file=task_config_path, local_rank=0) | ||
| config = GSConfig(args) | ||
| tasks = config.multi_tasks | ||
|
|
||
| # generate random node ids for nodes | ||
| nid0 = np.random.randint(10000, size=1000) * 10000 + np.arange(1000) | ||
| nid1 = np.random.randint(10000, size=1000) * 10000 + np.arange(1000) | ||
| nid0_str = nid0.astype('str') | ||
| nid1_str = nid1.astype('str') | ||
|
|
||
| nid0_map = IdMap(nid0_str) | ||
| nid1_map = IdMap(nid1_str) | ||
|
|
||
| mapping_subpath = "raw_id_mappings" | ||
| os.makedirs(os.path.join(graph_part_dir, mapping_subpath), exist_ok=True) | ||
| nid0_map.save(os.path.join(graph_part_dir, mapping_subpath, ntype0, "part-00000.parquet")) | ||
| nid1_map.save(os.path.join(graph_part_dir, mapping_subpath, ntype1, "part-00000.parquet")) | ||
|
|
||
| # generate faked edge results | ||
| src0 = nid0[np.random.randint(1000, size=2000)] | ||
| dst0 = nid1[np.random.randint(1000, size=2000)] | ||
| pred0 = np.stack((src0, dst0), axis=1) | ||
| src0_new, _ = nid0_map.map_id(src0.astype('str')) | ||
| dst0_new, _ = nid1_map.map_id(dst0.astype('str')) | ||
|
|
||
| src1 = nid0[np.random.randint(1000, size=2000)] | ||
| dst1 = nid1[np.random.randint(1000, size=2000)] | ||
| pred1 = np.stack((src1, dst1), axis=1) | ||
| src1_new, _ = nid0_map.map_id(src1.astype('str')) | ||
| dst1_new, _ = nid1_map.map_id(dst1.astype('str')) | ||
|
|
||
| meta_info = { | ||
| "format": "pytorch", | ||
| "world_size": 2, | ||
| "etypes": [etype0] | ||
| } | ||
|
|
||
| # For task 0 (edge classification task on etype0) | ||
| pred_output = os.path.join(save_prediction_path, tasks[0].task_id) | ||
| os.makedirs(pred_output) | ||
| meta_fname = os.path.join(pred_output, "result_info.json") | ||
| with open(meta_fname, 'w', encoding='utf-8') as f: | ||
| json.dump(meta_info, f, indent=4) | ||
|
|
||
| pred_output_etype0 = os.path.join(pred_output, "_".join(etype0)) | ||
| os.makedirs(pred_output_etype0) | ||
| th.save(th.tensor(pred0), os.path.join(pred_output_etype0, f"predict-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(src0_new), os.path.join(pred_output_etype0, f"src_nids-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(dst0_new), os.path.join(pred_output_etype0, f"dst_nids-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(pred1), os.path.join(pred_output_etype0, f"predict-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(src1_new), os.path.join(pred_output_etype0, f"src_nids-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(dst1_new), os.path.join(pred_output_etype0, f"dst_nids-{pad_file_index(1)}.pt")) | ||
|
|
||
| # generate faked edge results | ||
| src2 = nid1[np.random.randint(1000, size=2000)] | ||
| dst2 = nid0[np.random.randint(1000, size=2000)] | ||
| pred2 = np.stack((src2, dst2), axis=1) | ||
| src2_new, _ = nid1_map.map_id(src2.astype('str')) | ||
| dst2_new, _ = nid0_map.map_id(dst2.astype('str')) | ||
|
|
||
| src3 = nid1[np.random.randint(1000, size=2000)] | ||
| dst3 = nid0[np.random.randint(1000, size=2000)] | ||
| pred3 = np.stack((src3, dst3), axis=1) | ||
| src3_new, _ = nid1_map.map_id(src3.astype('str')) | ||
| dst3_new, _ = nid0_map.map_id(dst3.astype('str')) | ||
|
|
||
| meta_info = { | ||
| "format": "pytorch", | ||
| "world_size": 2, | ||
| "etypes": [etype1] | ||
| } | ||
|
|
||
| # For task 1 (edge classification task on etype0) | ||
| pred_output = os.path.join(save_prediction_path, tasks[1].task_id) | ||
| os.makedirs(pred_output) | ||
| meta_fname = os.path.join(pred_output, "result_info.json") | ||
| with open(meta_fname, 'w', encoding='utf-8') as f: | ||
| json.dump(meta_info, f, indent=4) | ||
|
|
||
| pred_output_etype1 = os.path.join(pred_output, "_".join(etype1)) | ||
| os.makedirs(pred_output_etype1) | ||
| th.save(th.tensor(pred2), os.path.join(pred_output_etype1, f"predict-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(src2_new), os.path.join(pred_output_etype1, f"src_nids-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(dst2_new), os.path.join(pred_output_etype1, f"dst_nids-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(pred3), os.path.join(pred_output_etype1, f"predict-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(src3_new), os.path.join(pred_output_etype1, f"src_nids-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(dst3_new), os.path.join(pred_output_etype1, f"dst_nids-{pad_file_index(1)}.pt")) | ||
|
|
||
|
|
||
| # generate faked edge results | ||
| nid0_0 = nid0[np.random.randint(1000, size=2000)] | ||
| pred0_0 = np.stack((nid0_0, nid0_0), axis=1) | ||
| nid0_new_0, _ = nid0_map.map_id(nid0_0.astype('str')) | ||
|
|
||
| nid0_1 = nid0[np.random.randint(1000, size=2000)] | ||
| pred0_1 = np.stack((nid0_1, nid0_1), axis=1) | ||
| nid0_new_1, _ = nid0_map.map_id(nid0_1.astype('str')) | ||
|
|
||
| meta_info = { | ||
| "format": "pytorch", | ||
| "world_size": 2, | ||
| "ntypes": [ntype0] | ||
| } | ||
| # For task 1 (edge classification task on etype0) | ||
| pred_output = os.path.join(save_prediction_path, tasks[2].task_id) | ||
| os.makedirs(pred_output) | ||
| meta_fname = os.path.join(pred_output, "result_info.json") | ||
| with open(meta_fname, 'w', encoding='utf-8') as f: | ||
| json.dump(meta_info, f, indent=4) | ||
| pred_output_ntype0 = os.path.join(pred_output, ntype0) | ||
|
|
||
| os.makedirs(pred_output_ntype0) | ||
| th.save(th.tensor(pred0_0), os.path.join(pred_output_ntype0, f"predict-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(nid0_new_0), os.path.join(pred_output_ntype0, f"predict_nids-{pad_file_index(0)}.pt")) | ||
|
|
||
| th.save(th.tensor(pred0_1), os.path.join(pred_output_ntype0, f"predict-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(nid0_new_1), os.path.join(pred_output_ntype0, f"predict_nids-{pad_file_index(1)}.pt")) | ||
|
|
||
| # Only part of nodes have node embeddings | ||
| # generate faked node embeddings | ||
| nid0_0 = nid0[np.random.randint(1000, size=250)] | ||
| nid1_0 = nid1[np.random.randint(1000, size=249)] | ||
| emb0_0 = np.stack((nid0_0, nid0_0), axis=1) | ||
| emb1_0 = np.stack((nid1_0, nid1_0), axis=1) | ||
| nid0_new_0, _ = nid0_map.map_id(nid0_0.astype('str')) | ||
| nid1_new_0, _ = nid1_map.map_id(nid1_0.astype('str')) | ||
|
|
||
| nid0_1 = nid0[np.random.randint(1000, size=250)] | ||
| nid1_1 = nid1[np.random.randint(1000, size=249)] | ||
| emb0_1 = np.stack((nid0_1, nid0_1), axis=1) | ||
| emb1_1 = np.stack((nid1_1, nid1_1), axis=1) | ||
| nid0_new_1, _ = nid0_map.map_id(nid0_1.astype('str')) | ||
| nid1_new_1, _ = nid1_map.map_id(nid1_1.astype('str')) | ||
|
|
||
| meta_info = { | ||
| "format": "pytorch", | ||
| "world_size": 2, | ||
| "emb_name": [ntype0, ntype1], | ||
| } | ||
| emb_output = save_embed_path | ||
| meta_fname = os.path.join(emb_output, "emb_info.json") | ||
| os.makedirs(emb_output) | ||
| with open(meta_fname, 'w', encoding='utf-8') as f: | ||
| json.dump(meta_info, f, indent=4) | ||
|
|
||
| emb_output_ntype0 = os.path.join(emb_output, ntype0) | ||
| emb_output_ntype1 = os.path.join(emb_output, ntype1) | ||
| os.makedirs(emb_output_ntype0) | ||
| os.makedirs(emb_output_ntype1) | ||
| th.save(th.tensor(emb0_0), os.path.join(emb_output_ntype0, f"embed-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(nid0_new_0), os.path.join(emb_output_ntype0, f"embed_nids-{pad_file_index(0)}.pt")) | ||
|
|
||
| th.save(th.tensor(emb1_0), os.path.join(emb_output_ntype1, f"embed-{pad_file_index(0)}.pt")) | ||
| th.save(th.tensor(nid1_new_0), os.path.join(emb_output_ntype1, f"embed_nids-{pad_file_index(0)}.pt")) | ||
|
|
||
| th.save(th.tensor(emb0_1), os.path.join(emb_output_ntype0, f"embed-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(nid0_new_1), os.path.join(emb_output_ntype0, f"embed_nids-{pad_file_index(1)}.pt")) | ||
|
|
||
| th.save(th.tensor(emb1_1), os.path.join(emb_output_ntype1, f"embed-{pad_file_index(1)}.pt")) | ||
| th.save(th.tensor(nid1_new_1), os.path.join(emb_output_ntype1, f"embed_nids-{pad_file_index(1)}.pt")) | ||
|
|
||
|
|
||
|
|
||
|
|
||
| if __name__ == '__main__': | ||
| argparser = argparse.ArgumentParser("Check multi task remapping") | ||
| argparser.add_argument("--output", type=str, required=True, | ||
| help="Path to save the generated data") | ||
|
|
||
| args = argparser.parse_args() | ||
|
|
||
| main(args) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters