314 ml-1m dim={8,16,32,64}实验结果

deipss · Jun 6, 2021 · 45e31f0 · 45e31f0
1 parent 99c1f63
commit 45e31f0
Show file tree

Hide file tree

Showing 9 changed files with 360 additions and 1 deletion.
diff --git a/experiments/test_bert_16_ml-1m_2021-06-05_0/config.json b/experiments/test_bert_16_ml-1m_2021-06-05_0/config.json
@@ -0,0 +1,74 @@
+{
+  "mode": "train",
+  "template": "train_bert",
+  "test_model_path": null,
+  "dataset_code": "ml-1m",
+  "min_rating": 0,
+  "min_uc": 5,
+  "min_sc": 0,
+  "split": "leave_one_out",
+  "dataset_split_seed": 98765,
+  "eval_set_size": 500,
+  "dataloader_code": "bert",
+  "dataloader_random_seed": 0.0,
+  "train_batch_size": 128,
+  "val_batch_size": 128,
+  "test_batch_size": 128,
+  "train_negative_sampler_code": "random",
+  "train_negative_sample_size": 0,
+  "train_negative_sampling_seed": 0,
+  "test_negative_sampler_code": "random",
+  "test_negative_sample_size": 100,
+  "test_negative_sampling_seed": 98765,
+  "trainer_code": "bert",
+  "device": "cuda",
+  "num_gpu": 1,
+  "device_idx": "0",
+  "optimizer": "Adam",
+  "lr": 0.001,
+  "weight_decay": 0,
+  "momentum": null,
+  "decay_step": 25,
+  "gamma": 1.0,
+  "num_epochs": 100,
+  "log_period_as_iter": 12800,
+  "metric_ks": [
+    1,
+    5,
+    10,
+    15,
+    20,
+    25,
+    30
+  ],
+  "best_metric": "NDCG@10",
+  "find_best_beta": false,
+  "total_anneal_steps": 2000,
+  "anneal_cap": 0.2,
+  "model_code": "bert",
+  "model_init_seed": 0,
+  "bert_max_len": 100,
+  "bert_num_items": null,
+  "bert_hidden_units": 16,
+  "bert_num_blocks": 2,
+  "bert_num_heads": 2,
+  "bert_dropout": 0.1,
+  "bert_mask_prob": 0.15,
+  "kernel_size": 3,
+  "stride": 2,
+  "dae_num_items": null,
+  "dae_num_hidden": 0,
+  "dae_hidden_dim": 600,
+  "dae_latent_dim": 16,
+  "dae_dropout": 0.5,
+  "vae_num_items": null,
+  "vae_num_hidden": 0,
+  "vae_hidden_dim": 600,
+  "vae_latent_dim": 16,
+  "vae_dropout": 0.5,
+  "experiment_dir": "experiments",
+  "experiment_description": "test",
+  "dim": 16,
+  "enable_lr_schedule": true,
+  "num_items": 3706
+}
diff --git a/experiments/test_bert_16_ml-1m_2021-06-05_0/logs/test_metrics.json b/experiments/test_bert_16_ml-1m_2021-06-05_0/logs/test_metrics.json
@@ -0,0 +1,16 @@
+{
+    "Recall@30": 0.84809027860562,
+    "NDCG@30": 0.39249670195082825,
+    "Recall@25": 0.8092447916666666,
+    "NDCG@25": 0.3844771229972442,
+    "Recall@20": 0.7559136288861433,
+    "NDCG@20": 0.3728274628520012,
+    "Recall@15": 0.6851671015222868,
+    "NDCG@15": 0.35611255280673504,
+    "Recall@10": 0.578125,
+    "NDCG@10": 0.32781057618558407,
+    "Recall@5": 0.4091796875,
+    "NDCG@5": 0.27322401634107035,
+    "Recall@1": 0.13069661458333334,
+    "NDCG@1": 0.13069661458333334
+}
diff --git a/experiments/test_bert_32_ml-1m_2021-06-05_0/config.json b/experiments/test_bert_32_ml-1m_2021-06-05_0/config.json
@@ -0,0 +1,74 @@
+{
+  "mode": "train",
+  "template": "train_bert",
+  "test_model_path": null,
+  "dataset_code": "ml-1m",
+  "min_rating": 0,
+  "min_uc": 5,
+  "min_sc": 0,
+  "split": "leave_one_out",
+  "dataset_split_seed": 98765,
+  "eval_set_size": 500,
+  "dataloader_code": "bert",
+  "dataloader_random_seed": 0.0,
+  "train_batch_size": 128,
+  "val_batch_size": 128,
+  "test_batch_size": 128,
+  "train_negative_sampler_code": "random",
+  "train_negative_sample_size": 0,
+  "train_negative_sampling_seed": 0,
+  "test_negative_sampler_code": "random",
+  "test_negative_sample_size": 100,
+  "test_negative_sampling_seed": 98765,
+  "trainer_code": "bert",
+  "device": "cuda",
+  "num_gpu": 1,
+  "device_idx": "0",
+  "optimizer": "Adam",
+  "lr": 0.001,
+  "weight_decay": 0,
+  "momentum": null,
+  "decay_step": 25,
+  "gamma": 1.0,
+  "num_epochs": 100,
+  "log_period_as_iter": 12800,
+  "metric_ks": [
+    1,
+    5,
+    10,
+    15,
+    20,
+    25,
+    30
+  ],
+  "best_metric": "NDCG@10",
+  "find_best_beta": false,
+  "total_anneal_steps": 2000,
+  "anneal_cap": 0.2,
+  "model_code": "bert",
+  "model_init_seed": 0,
+  "bert_max_len": 100,
+  "bert_num_items": null,
+  "bert_hidden_units": 32,
+  "bert_num_blocks": 2,
+  "bert_num_heads": 2,
+  "bert_dropout": 0.1,
+  "bert_mask_prob": 0.15,
+  "kernel_size": 3,
+  "stride": 2,
+  "dae_num_items": null,
+  "dae_num_hidden": 0,
+  "dae_hidden_dim": 600,
+  "dae_latent_dim": 32,
+  "dae_dropout": 0.5,
+  "vae_num_items": null,
+  "vae_num_hidden": 0,
+  "vae_hidden_dim": 600,
+  "vae_latent_dim": 32,
+  "vae_dropout": 0.5,
+  "experiment_dir": "experiments",
+  "experiment_description": "test",
+  "dim": 32,
+  "enable_lr_schedule": true,
+  "num_items": 3706
+}
diff --git a/experiments/test_bert_32_ml-1m_2021-06-05_0/logs/test_metrics.json b/experiments/test_bert_32_ml-1m_2021-06-05_0/logs/test_metrics.json
@@ -0,0 +1,16 @@
+{
+    "Recall@30": 0.87451171875,
+    "NDCG@30": 0.4707090798765421,
+    "Recall@25": 0.8492838541666666,
+    "NDCG@25": 0.46551593641440075,
+    "Recall@20": 0.8077256952722868,
+    "NDCG@20": 0.45642068050801754,
+    "Recall@15": 0.7513563372194767,
+    "NDCG@15": 0.4430973989268144,
+    "Recall@10": 0.6700303827722868,
+    "NDCG@10": 0.42155429534614086,
+    "Recall@5": 0.5216471354166666,
+    "NDCG@5": 0.3735509055356185,
+    "Recall@1": 0.21321614583333334,
+    "NDCG@1": 0.21321614583333334
+}
diff --git a/experiments/test_bert_64_ml-1m_2021-06-05_0/config.json b/experiments/test_bert_64_ml-1m_2021-06-05_0/config.json
@@ -0,0 +1,74 @@
+{
+  "mode": "train",
+  "template": "train_bert",
+  "test_model_path": null,
+  "dataset_code": "ml-1m",
+  "min_rating": 0,
+  "min_uc": 5,
+  "min_sc": 0,
+  "split": "leave_one_out",
+  "dataset_split_seed": 98765,
+  "eval_set_size": 500,
+  "dataloader_code": "bert",
+  "dataloader_random_seed": 0.0,
+  "train_batch_size": 128,
+  "val_batch_size": 128,
+  "test_batch_size": 128,
+  "train_negative_sampler_code": "random",
+  "train_negative_sample_size": 0,
+  "train_negative_sampling_seed": 0,
+  "test_negative_sampler_code": "random",
+  "test_negative_sample_size": 100,
+  "test_negative_sampling_seed": 98765,
+  "trainer_code": "bert",
+  "device": "cuda",
+  "num_gpu": 1,
+  "device_idx": "0",
+  "optimizer": "Adam",
+  "lr": 0.001,
+  "weight_decay": 0,
+  "momentum": null,
+  "decay_step": 25,
+  "gamma": 1.0,
+  "num_epochs": 100,
+  "log_period_as_iter": 12800,
+  "metric_ks": [
+    1,
+    5,
+    10,
+    15,
+    20,
+    25,
+    30
+  ],
+  "best_metric": "NDCG@10",
+  "find_best_beta": false,
+  "total_anneal_steps": 2000,
+  "anneal_cap": 0.2,
+  "model_code": "bert",
+  "model_init_seed": 0,
+  "bert_max_len": 100,
+  "bert_num_items": null,
+  "bert_hidden_units": 64,
+  "bert_num_blocks": 2,
+  "bert_num_heads": 2,
+  "bert_dropout": 0.1,
+  "bert_mask_prob": 0.15,
+  "kernel_size": 3,
+  "stride": 2,
+  "dae_num_items": null,
+  "dae_num_hidden": 0,
+  "dae_hidden_dim": 600,
+  "dae_latent_dim": 64,
+  "dae_dropout": 0.5,
+  "vae_num_items": null,
+  "vae_num_hidden": 0,
+  "vae_hidden_dim": 600,
+  "vae_latent_dim": 64,
+  "vae_dropout": 0.5,
+  "experiment_dir": "experiments",
+  "experiment_description": "test",
+  "dim": 64,
+  "enable_lr_schedule": true,
+  "num_items": 3706
+}
diff --git a/experiments/test_bert_64_ml-1m_2021-06-05_0/logs/test_metrics.json b/experiments/test_bert_64_ml-1m_2021-06-05_0/logs/test_metrics.json
@@ -0,0 +1,16 @@
+{
+    "Recall@30": 0.8938802083333334,
+    "NDCG@30": 0.5361959642420212,
+    "Recall@25": 0.8653428827722868,
+    "NDCG@25": 0.5303041978428761,
+    "Recall@20": 0.8298068580528101,
+    "NDCG@20": 0.5225369967520237,
+    "Recall@15": 0.7862413202722868,
+    "NDCG@15": 0.5122410276283821,
+    "Recall@10": 0.7253689244389534,
+    "NDCG@10": 0.49607940887411434,
+    "Recall@5": 0.6006401913861433,
+    "NDCG@5": 0.4556624299536149,
+    "Recall@1": 0.29161241340140504,
+    "NDCG@1": 0.29161241340140504
+}
diff --git a/experiments/test_bert_8_ml-1m_2021-06-05_0/config.json b/experiments/test_bert_8_ml-1m_2021-06-05_0/config.json
@@ -0,0 +1,73 @@
+{
+  "mode": "train",
+  "template": "train_bert",
+  "test_model_path": null,
+  "dataset_code": "ml-1m",
+  "min_rating": 0,
+  "min_uc": 5,
+  "min_sc": 0,
+  "split": "leave_one_out",
+  "dataset_split_seed": 98765,
+  "eval_set_size": 500,
+  "dataloader_code": "bert",
+  "dataloader_random_seed": 0.0,
+  "train_batch_size": 128,
+  "val_batch_size": 128,
+  "test_batch_size": 128,
+  "train_negative_sampler_code": "random",
+  "train_negative_sample_size": 0,
+  "train_negative_sampling_seed": 0,
+  "test_negative_sampler_code": "random",
+  "test_negative_sample_size": 100,
+  "test_negative_sampling_seed": 98765,
+  "trainer_code": "bert",
+  "device": "cuda",
+  "num_gpu": 1,
+  "device_idx": "0",
+  "optimizer": "Adam",
+  "lr": 0.001,
+  "weight_decay": 0,
+  "momentum": null,
+  "decay_step": 25,
+  "gamma": 1.0,
+  "num_epochs": 100,
+  "log_period_as_iter": 12800,
+  "metric_ks": [
+    1,
+    5,
+    10,
+    15,
+    20,
+    25,
+    30
+  ],
+  "best_metric": "NDCG@10",
+  "find_best_beta": false,
+  "total_anneal_steps": 2000,
+  "anneal_cap": 0.2,
+  "model_code": "bert",
+  "model_init_seed": 0,
+  "bert_max_len": 100,
+  "bert_num_items": null,
+  "bert_hidden_units": 8,
+  "bert_num_blocks": 2,
+  "bert_num_heads": 2,
+  "bert_dropout": 0.1,
+  "bert_mask_prob": 0.15,
+  "kernel_size": 3,
+  "stride": 2,
+  "dae_num_items": null,
+  "dae_num_hidden": 0,
+  "dae_hidden_dim": 600,
+  "dae_latent_dim": 8,
+  "dae_dropout": 0.5,
+  "vae_num_items": null,
+  "vae_num_hidden": 0,
+  "vae_hidden_dim": 600,
+  "vae_latent_dim": 8,
+  "vae_dropout": 0.5,
+  "experiment_dir": "experiments",
+  "experiment_description": "test",
+  "dim": 8,
+  "enable_lr_schedule": true
+}
diff --git a/experiments/test_bert_8_ml-1m_2021-06-05_0/logs/test_metrics.json b/experiments/test_bert_8_ml-1m_2021-06-05_0/logs/test_metrics.json
@@ -0,0 +1,16 @@
+{
+    "Recall@30": 0.8117404518028101,
+    "NDCG@30": 0.3501921445131302,
+    "Recall@25": 0.7603624140222868,
+    "NDCG@25": 0.33960138571759063,
+    "Recall@20": 0.6925998268028101,
+    "NDCG@20": 0.32480075148244697,
+    "Recall@15": 0.6105685768028101,
+    "NDCG@15": 0.30545171691725653,
+    "Recall@10": 0.5061848958333334,
+    "NDCG@10": 0.277898524577419,
+    "Recall@5": 0.3340386288861434,
+    "NDCG@5": 0.22245480647931495,
+    "Recall@1": 0.10590277782951792,
+    "NDCG@1": 0.10590277782951792
+}
diff --git a/templates.py b/templates.py
@@ -54,7 +54,7 @@ def set_template(args):
     elif args.template.startswith('train_bert'):
         args.mode = 'train'
         # args.dataset_code = 'ml-' + input('Input 1 for ml-1m, 20 for ml-20m: ') + 'm'
-        args.dataset_code = 'magazine'
+        args.dataset_code = 'ml-1m'
         args.min_rating = 0 if args.dataset_code == 'ml-1m' else 4
         args.min_uc = 5
         args.min_sc = 0