Fix seed of simulation

diploma_thesis/breakdown/dynamic.py

@@ -1,4 +1,4 @@
-
+import torch
 
 from environment import Breakdown, Machine
 from typing import Dict
@@ -18,6 +18,10 @@ def __init__(self, configuration: Configuration):
 
         self.configuration = configuration
 
+    def connect(self, generator: torch.Generator):
+        self.configuration.arrival.connect(generator)
+        self.configuration.duration.connect(generator)
+
     def sample_next_breakdown_time(self, machine: Machine):
         return self.configuration.arrival.sample((1,))
 

diploma_thesis/breakdown/no.py

@@ -1,10 +1,14 @@
+import torch
 
 from environment import Breakdown, Machine
 from typing import Dict
 
 
 class No(Breakdown):
 
+    def connect(self, generator: torch.Generator):
+        pass
+
     def sample_next_breakdown_time(self, machine: Machine):
         return float('inf')
 

diploma_thesis/configuration/jsp.yml

@@ -1,50 +1,144 @@
+jsp: &jsp
+  name: 'model'
+  output_dir: 'tmp/jsp'
+  log_stdout: False
+
+  machine_agent:
+    kind: 'mod'
+    parameters:
+      base_path: 'configuration/mods/machine_agent/model.yml'
+      mods: [ ]
+
+  work_center_agent:
+    kind: 'static'
+    parameters:
+      model:
+        kind: 'static'
+        parameters:
+          rule: 'et'
+      encoder:
+        kind: 'plain'
+
+  tape:
+    machine_reward:
+      kind: 'global_tardiness'
+      parameters:
+        span: 256
+
+    work_center_reward:
+      kind: 'no'
+
+  simulator:
+    kind: 'td'
+
+  run:
+    kind: 'mod'
+    parameters:
+      base_path: 'configuration/mods/run/jsp/run.yml'
+      mods: [ ]
+
+rl_agents: &rl_agents
+  [
+    'rl/trainer/dqn.yml',
+    'rl/trainer/ddqn.yml'
+  ]
+
+memory: &memory
+  [
+    'memory/plain.yml',
+    'memory/prioritized.yml'
+  ]
+
+loss: &loss
+  [
+#    'loss/huber.yml',
+    'loss/mse.yml',
+#    'loss/smooth_l1.yml'
+  ]
+
+agent: &agent
+  [
+    'rl/agent/single.yml',
+    'rl/agent/multi_agent.yml',
+    'rl/agent/centralized.yml'
+  ]
+
+optimizer: &optimizer
+  [
+    'optimizer/adam.yml',
+#    'optimizer/sgd.yml'
+  ]
+
+action_set: &action_set
+  [
+    'model/rules/all.yml',
+    'model/rules/marl.yml'
+  ]
+
+reward: &reward
+  - kind: 'global_tardiness'
+  - kind: 'global_decomposed_tardiness'
+  - kind: 'surrogate_tardiness'
+
+run: &run
+  mods:
+#    - [ 'multi.yml' ]
+    - [ 'multi.yml', 'concurrent.yml' ]
+#    - [ 'util_70.yml' ]
+#    - [ 'util_80.yml' ]
+#    - [ 'util_90.yml' ]
 
 task:
   kind: 'multi_task'
   n_workers: 8
   debug: False
 
   tasks:
+    # AS
     - kind: 'multi_value'
       parameters:
         base:
-          name: 'model'
-          output_dir: 'tmp/jsp'
-          log_stdout: False
-
+          *jsp
+        values:
           machine_agent:
-            kind: 'mod'
-            parameters:
-              base_path: 'configuration/mods/machine_agent/model.yml'
-              mods: [ ]
-
-          work_center_agent:
-            kind: 'static'
             parameters:
-              model:
-                kind: 'static'
-                parameters:
-                  rule: 'et'
-              encoder:
-                kind: 'plain'
+              mods:
+                # Reserved keywords to build GridFactory resulting a list
+                __factory__:
+                  # RL Trainers
+                  - *rl_agents
+                  # RL Agent Kind
+                  - *agent
+                  # Memory
+                  - *memory
+                  # Loss
+                  - *loss
+                  # Optimizer
+                  - *optimizer
+                  # State Encoding
+                  - [
+                    'encoding/indirect.yml',
+                  ]
+                  # Action Set
+                  - *action_set
+                  # Model
+                  - [
+                    'model/model/marl_as.yml'
+                  ]
 
           tape:
             machine_reward:
-              kind: 'global_tardiness'
-              parameters:
-                span: 256
-
-            work_center_reward:
-              kind: 'no'
-
-          simulator:
-            kind: 'td'
+              *reward
 
           run:
-            kind: 'mod'
-            parameters:
-              base_path: 'configuration/mods/run/jsp/run.yml'
-              mods: [ ]
+            paramemeters:
+              *run
+
+    # MR
+    - kind: 'multi_value'
+      parameters:
+        base:
+          *jsp
 
         values:
           machine_agent:
@@ -53,61 +147,30 @@ task:
                 # Reserved keywords to build GridFactory resulting a list
                 __factory__:
                   # RL Trainers
-                  - [
-                    'rl/trainer/dqn.yml',
-                    'rl/trainer/ddqn.yml'
-                  ]
+                  - *rl_agents
                   # RL Agent Kind
-                  - [
-                    'rl/agent/single.yml',
-#                   'rl/agent/multi_agent.yml',
-#                   'rl/agent/centralized.yml'
-                  ]
+                  - *agent
                   # Memory
-                  - [
-                    'memory/plain.yml',
-                    'memory/prioritized.yml'
-                  ]
+                  - *memory
                   # Loss
-                  - [
-                    'loss/huber.yml',
-                    'loss/mse.yml',
-                    'loss/smooth_l1.yml'
-                  ]
+                  - *loss
                   # Optimizer
-                  - [
-                    'optimizer/adam.yml',
-                    'optimizer/sgd.yml'
-                  ]
+                  - *optimizer
                   # State Encoding
                   - [
-                    'encoding/indirect.yml',
                     'encoding/mr.yml'
                   ]
                   # Action Set
-                  - [
-                    'model/rules/all.yml',
-                    'model/rules/marl.yml'
-                  ]
+                  - *action_set
                   # Model
                   - [
-                    'model/model/marl_as.yml',
                     'model/model/marl_mr.yml'
                   ]
 
           tape:
             machine_reward:
-              - kind: 'global_tardiness'
-              - kind: 'global_decomposed_tardiness'
-              - kind: 'surrogate_tardiness'
+              *reward
 
           run:
             parameters:
-              mods:
-#                - ['test.yml']
-                - [ 'multi.yml' ]
-#                - [ 'multi.yml', 'concurrent.yml' ]
-#                - [ 'util_70.yml' ]
-#                - [ 'util_80.yml' ]
-#                - [ 'util_90.yml' ]
-
+              *run

diploma_thesis/environment/breakdown.py

@@ -1,7 +1,12 @@
+import torch
+from abc import ABCMeta
 import environment
 
 
-class Breakdown:
+class Breakdown(metaclass=ABCMeta):
+
+    def connect(self, generator: torch.Generator):
+        pass
 
     def sample_next_breakdown_time(self, machine: 'environment.Machine') -> float:
         pass

diploma_thesis/environment/job_sampler.py

@@ -14,6 +14,10 @@ def __init__(self, problem: Configuration, environment: simpy.Environment):
         self.problem = problem
         self.environment = environment
 
+    def connect(self, generator: torch.Generator):
+        pass
+
+    @abstractmethod
     def number_of_jobs(self):
         return
 

diploma_thesis/environment/shop_floor.py

@@ -107,7 +107,11 @@ def __init__(self, id: int, name: str, configuration: Configuration, logger: log
         self.did_finish_simulation_event = configuration.environment.event()
 
     def simulate(self):
-        torch.manual_seed(self.configuration.problem.seed)
+        generator = torch.Generator()
+        generator.manual_seed(self.configuration.problem.seed)
+
+        self.configuration.breakdown.connect(generator)
+        self.configuration.sampler.connect(generator)
 
         self.history.with_started_at(self.configuration.environment.now)
         self.delegate.did_start_simulation(context=self.__make_context__())

diploma_thesis/environment/statistics.py

@@ -7,7 +7,6 @@
 
 import pandas as pd
 import torch
-import torchsnapshot
 from tensordict import TensorDict
 
 import environment

diploma_thesis/job_sampler/dynamic/builder.py

@@ -22,6 +22,7 @@ def sample(initial_work_center_idx: int):
             return sampler.sample([self.problem.work_center_count])
 
         self.job_sampler._step_sampler = sample
+        self.job_sampler.store(sampler)
 
         return self
 
@@ -33,6 +34,7 @@ def sample(shape: Tuple[int]) -> torch.FloatTensor:
             return times
 
         self.job_sampler._processing_time_sampler = sample
+        self.job_sampler.store(sampler)
 
         return self
 
@@ -47,6 +49,7 @@ def sample(job: Job, moment: torch.FloatTensor) -> torch.FloatTensor:
             return torch.round(weight * tightness + moment)
 
         self.job_sampler._due_time_sampler = sample
+        self.job_sampler.store(sampler)
 
         return self
 
@@ -56,6 +59,7 @@ def sample() -> torch.FloatTensor:
 
         self.job_sampler.arrival_time_sampler = sample
         self.job_sampler._number_of_jobs = n_jobs
+        self.job_sampler.store(sampler)
 
         return self
 

diploma_thesis/job_sampler/dynamic/job_sampler.py

@@ -5,6 +5,7 @@
 import torch
 
 from environment import Configuration, Job, JobSampler as JSampler
+from sampler import Sampler
 
 
 class JobSampler(JSampler):
@@ -16,8 +17,13 @@ def __init__(self, problem: Configuration, environment: simpy.Environment):
         self._processing_time_sampler: Callable[[Tuple[int]], torch.FloatTensor] = None
         self._step_sampler: Callable[[int], torch.LongTensor] = None
         self._due_time_sampler: Callable[['Job', int], torch.FloatTensor] = None
+        self.samplers: [Sampler] = []
         self.arrival_time_sampler = None
 
+    def connect(self, generator: torch.Generator):
+        for sampler in self.samplers:
+            sampler.connect(generator)
+
     def number_of_jobs(self):
         return self._number_of_jobs
 
@@ -39,3 +45,6 @@ def sample(self, job_id: int, initial_work_center_idx: int, moment: torch.FloatT
 
     def sample_next_arrival_time(self) -> torch.FloatTensor:
         return self.arrival_time_sampler()
+
+    def store(self, sampler: Sampler):
+        self.samplers += [sampler]

diploma_thesis/sampler/constant.py

@@ -6,6 +6,8 @@
 class Constant(NumericSampler):
 
     def __init__(self, value: float):
+        super().__init__()
+
         self.value = value
 
     def sample(self, shape):