add trained models

job.py

@@ -91,6 +91,9 @@ def __init__(self, line = "0        0      0    0   0     0    0   0  0 0  0   0
     def __eq__(self, other):
         return self.job_id == other.job_id
 
+    def __lt__(self, other):
+        return self.job_id < other.job_id
+
     def __str__(self):
         return "J["+str(self.job_id)+"]-["+str(self.request_number_of_processors)+"]-["+str(self.submit_time)+"]-["+str(self.request_time)+"]"
     def __feature__(self):

trained_models/Lublin256new-nobackfil-8features/Lublin256new-nobackfil-8features_s0/config.json

@@ -0,0 +1,42 @@
+{
+    "ac_kwargs":	{},
+    "attn":	0,
+    "clip_ratio":	0.2,
+    "epochs":	4000,
+    "exp_name":	"Lublin256new-nobackfil-8features",
+    "gamma":	1,
+    "lam":	0.97,
+    "logger":	{
+        "<spinup.utils.logx.EpochLogger object at 0x7f2e67b55ed0>":	{
+            "epoch_dict":	{},
+            "exp_name":	"Lublin256new-nobackfil-8features",
+            "first_row":	true,
+            "log_current_row":	{},
+            "log_headers":	[],
+            "output_dir":	"/home/ddai/deep-batch-scheduler/./data/logs/Lublin256new-nobackfil-8features/Lublin256new-nobackfil-8features_s0",
+            "output_file":	{
+                "<_io.TextIOWrapper name='/home/ddai/deep-batch-scheduler/./data/logs/Lublin256new-nobackfil-8features/Lublin256new-nobackfil-8features_s0/progress.txt' mode='w' encoding='UTF-8'>":	{
+                    "mode":	"w"
+                }
+            }
+        }
+    },
+    "logger_kwargs":	{
+        "exp_name":	"Lublin256new-nobackfil-8features",
+        "output_dir":	"/home/ddai/deep-batch-scheduler/./data/logs/Lublin256new-nobackfil-8features/Lublin256new-nobackfil-8features_s0"
+    },
+    "max_ep_len":	1000,
+    "model_path":	"./data/lublin_256.schd",
+    "pi_lr":	0.0003,
+    "pre_trained":	0,
+    "save_freq":	10,
+    "seed":	0,
+    "shuffle":	0,
+    "target_kl":	0.01,
+    "train_pi_iters":	80,
+    "train_v_iters":	80,
+    "trained_model":	null,
+    "traj_per_epoch":	100,
+    "vf_lr":	0.001,
+    "workload_file":	"/home/ddai/deep-batch-scheduler/./data/lublin_256_new2"
+}

trained_models/Readme.md

@@ -0,0 +1,19 @@
+# Trained Models
+
+Models:
+
+lublin256-nobackfil 
+
+Lublin256new-nobackfil-8features 
+
+hpc2n-10k 
+
+sdsc_sp2-10k
+
+
+
+# Model Performance on different seeds and sequence length
+
+| Models        | lublin256-nobackfil                      | Lublin256new-nobackfil-8features            | hpc2n-10k                            | sdsc_sp2-10k                            |
+|---------------|------------------------------------------|---------------------------------------------|--------------------------------------|-----------------------------------------|
+| sequence=1024 | ![alt](resources/lublin256_training.png) | ![alt](resources/lublin256new_training.png) | ![alt](resources/hpc2n_training.png) | ![alt](resources/sdsc-sp2_training.png) |

trained_models/hpc2n-10k/hpc2n-10k_s4/config.json

@@ -0,0 +1,46 @@
+{
+    "ac_kwargs":	{},
+    "attn":	0,
+    "backfil":	0,
+    "batch_job_slice":	10000,
+    "clip_ratio":	0.2,
+    "epochs":	4000,
+    "exp_name":	"hpc2n-10k",
+    "gamma":	1,
+    "lam":	0.97,
+    "logger":	{
+        "<spinup.utils.logx.EpochLogger object at 0x7fd26db44d90>":	{
+            "epoch_dict":	{},
+            "exp_name":	"hpc2n-10k",
+            "first_row":	true,
+            "log_current_row":	{},
+            "log_headers":	[],
+            "output_dir":	"/users/dzhang16/deep-batch-scheduler/./data/logs/hpc2n-10k/hpc2n-10k_s4",
+            "output_file":	{
+                "<_io.TextIOWrapper name='/users/dzhang16/deep-batch-scheduler/./data/logs/hpc2n-10k/hpc2n-10k_s4/progress.txt' mode='w' encoding='UTF-8'>":	{
+                    "mode":	"w"
+                }
+            }
+        }
+    },
+    "logger_kwargs":	{
+        "exp_name":	"hpc2n-10k",
+        "output_dir":	"/users/dzhang16/deep-batch-scheduler/./data/logs/hpc2n-10k/hpc2n-10k_s4"
+    },
+    "max_ep_len":	1000,
+    "model_path":	"./data/lublin_256.schd",
+    "pi_lr":	0.0003,
+    "pre_trained":	0,
+    "save_freq":	10,
+    "score_type":	0,
+    "seed":	4,
+    "shuffle":	0,
+    "skip":	0,
+    "target_kl":	0.01,
+    "train_pi_iters":	80,
+    "train_v_iters":	80,
+    "trained_model":	null,
+    "traj_per_epoch":	100,
+    "vf_lr":	0.001,
+    "workload_file":	"/users/dzhang16/deep-batch-scheduler/data/HPC2N-2002-2.2-cln.swf"
+}

trained_models/lublin256-nobackfil/lublin256-nobackfil_s0/config.json

@@ -0,0 +1,42 @@
+{
+    "ac_kwargs":	{},
+    "attn":	0,
+    "clip_ratio":	0.2,
+    "epochs":	4000,
+    "exp_name":	"\u201clublin256-nobackfil\u201d",
+    "gamma":	1,
+    "lam":	0.97,
+    "logger":	{
+        "<spinup.utils.logx.EpochLogger object at 0x7fcde2ebcd50>":	{
+            "epoch_dict":	{},
+            "exp_name":	"\u201clublin256-nobackfil\u201d",
+            "first_row":	true,
+            "log_current_row":	{},
+            "log_headers":	[],
+            "output_dir":	"/home/ddai/deep-batch-scheduler/./data/logs/\u201clublin256-nobackfil\u201d/\u201clublin256-nobackfil\u201d_s0",
+            "output_file":	{
+                "<_io.TextIOWrapper name='/home/ddai/deep-batch-scheduler/./data/logs/\u201clublin256-nobackfil\u201d/\u201clublin256-nobackfil\u201d_s0/progress.txt' mode='w' encoding='UTF-8'>":	{
+                    "mode":	"w"
+                }
+            }
+        }
+    },
+    "logger_kwargs":	{
+        "exp_name":	"\u201clublin256-nobackfil\u201d",
+        "output_dir":	"/home/ddai/deep-batch-scheduler/./data/logs/\u201clublin256-nobackfil\u201d/\u201clublin256-nobackfil\u201d_s0"
+    },
+    "max_ep_len":	1000,
+    "model_path":	"./data/lublin_256.schd",
+    "pi_lr":	0.0003,
+    "pre_trained":	0,
+    "save_freq":	10,
+    "seed":	0,
+    "shuffle":	0,
+    "target_kl":	0.01,
+    "train_pi_iters":	80,
+    "train_v_iters":	80,
+    "trained_model":	null,
+    "traj_per_epoch":	100,
+    "vf_lr":	0.001,
+    "workload_file":	"/home/ddai/deep-batch-scheduler/./data/lublin_256.swf"
+}