homa 0.3.13__py3-none-any.whl → 0.3.19__py3-none-any.whl

homa/core/concerns/TracksTime.py

@@ -0,0 +1,7 @@
+class TracksTime:
+    def __init__(self):
+        super().__init__()
+        self.t = 0
+
+    def tick(self):
+        self.t += 1

homa/core/concerns/__init__.py

@@ -1 +1,2 @@
 from .MovesNetworkToDevice import MovesNetworkToDevice
+from .TracksTime import TracksTime

homa/ensemble/concerns/CalculatesMetricNecessities.py

@@ -3,8 +3,8 @@ from ...device import get_device
 
 
 class CalculatesMetricNecessities:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     @torch.no_grad()
     def metric_necessities(self, dataloader):

homa/ensemble/concerns/PredictsProbabilities.py

@@ -3,8 +3,8 @@ from .ReportsLogits import ReportsLogits
 
 
 class PredictsProbabilities(ReportsLogits):
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     def predict(self, x: torch.Tensor) -> torch.Tensor:
         logits = self.logits(x)

homa/ensemble/concerns/ReportsClassificationMetrics.py

@@ -10,4 +10,5 @@ class ReportsClassificationMetrics(
     ReportsEnsembleF1,
     ReportsEnsembleKappa,
 ):
-    pass
+    def __init__(self):
+        super().__init__()

homa/ensemble/concerns/ReportsEnsembleAccuracy.py

@@ -3,8 +3,8 @@ from torch.utils.data import DataLoader
 
 
 class ReportsEnsembleAccuracy:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     def accuracy(self, dataloader: DataLoader) -> float:
         predictions, labels = self.metric_necessities(dataloader)

homa/ensemble/concerns/ReportsEnsembleF1.py

@@ -2,8 +2,8 @@ from sklearn.metrics import f1_score as f1
 
 
 class ReportsEnsembleF1:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     def f1(self) -> float:
         predictions, labels = self.metric_necessities()

homa/ensemble/concerns/ReportsEnsembleKappa.py

@@ -2,8 +2,8 @@ from sklearn.metrics import cohen_kappa_score as kappa
 
 
 class ReportsEnsembleKappa:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     def accuracy(self) -> float:
         predictions, labels = self.metric_necessities()

homa/ensemble/concerns/ReportsEnsembleSize.py

@@ -1,6 +1,6 @@
 class ReportsEnsembleSize:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     @property
     def size(self):

homa/ensemble/concerns/ReportsLogits.py

@@ -2,11 +2,14 @@ import torch
 
 
 class ReportsLogits:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
-    def logits(self, x: torch.Tensor) -> torch.Tensor:
-        batch_size = x.shape[0]
+    def logits_average(self, x: torch.Tensor) -> torch.Tensor:
+        return self.logits_sim(x) / len(self.factories)
+
+    def logits_sum(self, x: torch.Tensor) -> torch.Tensor:
+        batch_size = x.size(0)
         logits = torch.zeros((batch_size, self.num_classes))
         for factory, weight in zip(self.factories, self.weights):
             model = factory(num_classes=self.num_classes)
@@ -14,6 +17,22 @@ class ReportsLogits:
             logits += model(x)
         return logits
 
+    def check_aggregation_strategy(self, aggregation: str):
+        if aggregation not in ["mean", "average", "sum"]:
+            raise ValueError(
+                f"Ensemble aggregation strategy must be in [mean, average, sum], but found {aggregation}."
+            )
+
+    def logits(self, x: torch.Tensor, aggregation: str = "mean") -> torch.Tensor:
+        self.check_aggregation_strategy(aggregation=aggregation)
+        logits_handlers = {
+            "mean": self.logits_average,
+            "average": self.logits_average,
+            "sum": self.logits_sum,
+        }
+        handler = logits_handlers.get(aggregation)
+        return handler(x)
+
     @torch.no_grad()
     def logits_(self, *args, **kwargs):
         return self.logits(*args, **kwargs)

homa/ensemble/concerns/SavesEnsembleModels.py

@@ -1,6 +1,6 @@
 class SavesEnsembleModels:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
 
     def save(self):
         self.save_factories()

homa/ensemble/concerns/StoresModels.py

@@ -5,8 +5,8 @@ from ...vision import Model
 
 
 class StoresModels:
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+    def __init__(self):
+        super().__init__()
         self.factories: List[Type[torch.nn.Module]] = []
         self.weights: List[OrderedDict] = []
 

homa/rl/SoftActorCritic.py

@@ -1,14 +1,15 @@
 from .sac import SoftActor, SoftCritic
 from .buffers import SoftActorCriticBuffer
+from ..core.concerns import TracksTime
 
 
-class SoftActorCritic:
+class SoftActorCritic(TracksTime):
     def __init__(
         self,
         state_dimension: int,
         action_dimension: int,
         hidden_dimension: int = 256,
-        buffer_capacity: int = 1_000_000,
+        buffer_capacity: int = 100_000,
         batch_size: int = 256,
         actor_lr: float = 0.0002,
         critic_lr: float = 0.0003,
@@ -19,10 +20,13 @@ class SoftActorCritic:
         gamma: float = 0.99,
         min_std: float = -20,
         max_std: float = 2,
-        warmup: int = 10_000,
+        warmup: int = 20_000,
     ):
+        super().__init__()
+
         self.batch_size: int = batch_size
         self.warmup: int = warmup
+        self.tau: float = tau
 
         self.actor = SoftActor(
             state_dimension=state_dimension,
@@ -40,7 +44,6 @@ class SoftActorCritic:
             hidden_dimension=hidden_dimension,
             lr=critic_lr,
             weight_decay=critic_decay,
-            tau=tau,
             gamma=gamma,
             alpha=alpha,
         )
@@ -61,4 +64,4 @@ class SoftActorCritic:
             actor=self.actor,
         )
         self.actor.train(states=data.states, critic_network=self.critic.network)
-        self.critic.update()
+        self.critic.update(tau=self.tau)

homa/rl/buffers/SoftActorCriticBuffer.py

@@ -33,6 +33,10 @@ class SoftActorCriticBuffer(Buffer):
         terminations = numpy.array(terminations)
         probabilities = numpy.array(probabilities)
 
+        # add one dimension to both rewards and terminations
+        rewards = numpy.expand_dims(rewards, axis=-1)
+        terminations = numpy.expand_dims(terminations, axis=-1)
+
         if as_tensor:
             states = torch.from_numpy(states).float()
             actions = torch.from_numpy(actions).float()

homa/rl/sac/SoftActor.py

@@ -29,6 +29,7 @@ class SoftActor:
         )
 
     def train(self, states: torch.Tensor, critic_network: torch.nn.Module):
+        self.network.train()
         self.optimizer.zero_grad()
         loss = self.loss(states=states, critic_network=critic_network)
         loss.backward()
@@ -44,7 +45,7 @@ class SoftActor:
 
     def process_state(self, state: numpy.ndarray | torch.Tensor) -> torch.Tensor:
         if isinstance(state, numpy.ndarray):
-            state = torch.from_numpy(state)
+            state = torch.from_numpy(state).float()
 
         if state.ndim < 2:
             state = state.unsqueeze(0)
@@ -64,6 +65,6 @@ class SoftActor:
         action = torch.tanh(pre_tanh)
 
         probabilities = distribution.log_prob(pre_tanh).sum(dim=1, keepdim=True)
-        correction = torch.log(1 - action.pow(2) + 1e-6).sum(dim=1, keepdim=True)
+        probabilities -= torch.log(1 - action.pow(2) + 1e-6).sum(dim=1, keepdim=True)
 
-        return action, probabilities - correction
+        return action, probabilities

homa/rl/sac/SoftCritic.py

@@ -13,11 +13,9 @@ class SoftCritic:
         action_dimension: int,
         lr: float,
         weight_decay: float,
-        tau: float,
         gamma: float,
         alpha: float,
     ):
-        self.tau: float = tau
         self.gamma: float = gamma
         self.alpha: float = alpha
 
@@ -48,6 +46,7 @@ class SoftCritic:
         next_states: torch.Tensor,
         actor: SoftActor,
     ):
+        self.network.train()
         self.optimizer.zero_grad()
         loss = self.loss(
             states=states,
@@ -93,6 +92,6 @@ class SoftCritic:
         entropy_q = q - self.alpha * next_probabilities
         return rewards + self.gamma * termination_mask * entropy_q
 
-    def update(self):
-        soft_update(network=self.network.alpha, target=self.target.alpha)
-        soft_update(network=self.network.beta, target=self.target.beta)
+    def update(self, tau: float):
+        soft_update(network=self.network.alpha, target=self.target.alpha, tau=tau)
+        soft_update(network=self.network.beta, target=self.target.beta, tau=tau)

homa/rl/utils.py

@@ -2,6 +2,6 @@ import torch
 
 
 @torch.no_grad()
-def soft_update(self, network: torch.nn.Module, target: torch.nn.Module):
+def soft_update(network: torch.nn.Module, target: torch.nn.Module, tau: float):
     for s, t in zip(network.parameters(), target.parameters()):
-        t.data.copy_(self.tau * s.data + (1 - self.tau) * t.data)
+        t.data.copy_(tau * s.data + (1 - tau) * t.data)

{homa-0.3.13.dist-info → homa-0.3.19.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: homa
-Version: 0.3.13
+Version: 0.3.19
 Summary: A curated list of machine learning and deep learning helpers.
 Author-email: Taha Shieenavaz <tahashieenavaz@gmail.com>
 Requires-Python: >=3.7

{homa-0.3.13.dist-info → homa-0.3.19.dist-info}/RECORD

@@ -73,20 +73,21 @@ homa/cli/namespaces/MakeNamespace.py,sha256=5G6LHk3lDkXROz7uq4jYE0DyO_V7JvnhJ33I
 homa/cli/namespaces/__init__.py,sha256=zAKUGPH4wcacxfH5Qvidp-uOuHdfzhan6kvVI6eMKA8,84
 homa/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 homa/core/concerns/MovesNetworkToDevice.py,sha256=OPMvO7scsM6NNy_fM0cJdkRdoVc-b2j6l4bz88cBif0,348
-homa/core/concerns/__init__.py,sha256=O9OXMIMYrkIgp11lAyEv-OgT3Wq0IvNdDVZr2bOmpQU,55
+homa/core/concerns/TracksTime.py,sha256=atg7iUH5HKqKJd03s9eHsl18iUO_4fzxuYmXgNtqSBQ,129
+homa/core/concerns/__init__.py,sha256=6jL3_kiqmmMs8BV789ZBbwEYNQNAhq1otVOrDJJrSXo,90
 homa/ensemble/Ensemble.py,sha256=mrqwbEm8OtiBmEgKuO6RzO1V8v80vrQFIJ4WHl8Yqgk,356
 homa/ensemble/__init__.py,sha256=1pk2W-NbgfDFh9WLKZVLUk2E3PTjVZ5Bap9dQEnrs9o,31
 homa/ensemble/utils.py,sha256=nn6eAgGW7ZafjjOVJWzGUWE0XYeyJAOMNEHm-lHxd6A,200
-homa/ensemble/concerns/CalculatesMetricNecessities.py,sha256=QccROg_FOp_X2T_lZDg8p1DMZhPYdO-7aEdnebRXMsY,825
-homa/ensemble/concerns/PredictsProbabilities.py,sha256=7rmI66DzE7-QGoJgZEk-9fu5YQvJW-4ZnMn_dWEEhqU,440
-homa/ensemble/concerns/ReportsClassificationMetrics.py,sha256=bg__cdCKp2U1H9qN1aOJH4BoX98oIvt8XaPDGApJhSM,395
-homa/ensemble/concerns/ReportsEnsembleAccuracy.py,sha256=AX5X3VGOm7DfdonW0N7FFgUwEr7wnsojRSVEULEii7c,380
-homa/ensemble/concerns/ReportsEnsembleF1.py,sha256=hdtdCQrWaFJNUn1KP9cAmi_q_EA4FYnpkBMlYLjzRZg,296
-homa/ensemble/concerns/ReportsEnsembleKappa.py,sha256=ZRbtrFCTD84EDql6ZL1xeWtTLFxpO5Y5tQaUlR6_0jw,300
-homa/ensemble/concerns/ReportsEnsembleSize.py,sha256=eIweQHpLcfGnNLwiMuTho-9rDgxV0xXGHPTOaEOABzw,240
-homa/ensemble/concerns/ReportsLogits.py,sha256=sJZGJwTISZo2DFmJbI5zqhrt7CblNi09iGn1zaEk-ro,593
-homa/ensemble/concerns/SavesEnsembleModels.py,sha256=VIXT9wJ8FiCspIvI2-F4WPa6mBBe9SWvMLFyad3TgRg,275
-homa/ensemble/concerns/StoresModels.py,sha256=uAYbdUtadZsAJ9-Fj4jJFLWC23qfiXKo1mBm6-PZkN4,963
+homa/ensemble/concerns/CalculatesMetricNecessities.py,sha256=HgrLbz8O9grGZZ0LG82Au5lZwq2D1zixDRjegM4f8Wk,793
+homa/ensemble/concerns/PredictsProbabilities.py,sha256=WWUaNXQxCJQ_NrLgeTdw0OXEsDb5xU7899_2d3Pzaoc,408
+homa/ensemble/concerns/ReportsClassificationMetrics.py,sha256=S9IBH6O7dmHhQ4Mxf5c7JFirOsPokKZUyqfOmED13mM,437
+homa/ensemble/concerns/ReportsEnsembleAccuracy.py,sha256=UuaPQ7v2sCaWuo3xa4PaTZzyjciRhcIluhnt6Zla2Fo,348
+homa/ensemble/concerns/ReportsEnsembleF1.py,sha256=aXKBK2-dTB133Rjg-X2a4Khb1sTdxuffXukcSMZlkzM,264
+homa/ensemble/concerns/ReportsEnsembleKappa.py,sha256=AkWTVGuCeIanDusNdtJOHwTgSEh5RJeWwLTQJJqSEKE,268
+homa/ensemble/concerns/ReportsEnsembleSize.py,sha256=lRyHIrK_zr7pE5RlwuNLIkqXoEMoVhNmT1nYDgCaNVI,208
+homa/ensemble/concerns/ReportsLogits.py,sha256=yZobLvxPL6ep70uMFIEtz5-l4rlfaG7m9mti6jJD1E8,1338
+homa/ensemble/concerns/SavesEnsembleModels.py,sha256=d1DcZnzfJABEfxcnYy5tV9N7YOghzO_ZdCdU80VTcno,243
+homa/ensemble/concerns/StoresModels.py,sha256=dg-xP1C4A9K8DrUTnR4VfqWU9iNAdS_0DlQcRThDka8,931
 homa/ensemble/concerns/__init__.py,sha256=IF5mHIgzCuCpA2EmpkctbjAr0kYW4P96v7RffK2V_iQ,548
 homa/graph/GraphAttention.py,sha256=oPXuc1s-3BXwGkHuomEIxnOcZSRBbL8b8fO0432RdDo,478
 homa/graph/__init__.py,sha256=NCtMUB-awe9UvkwDYqWXxTAZ1RW-AwSW1DD9X_kFkD0,43
@@ -99,13 +100,13 @@ homa/loss/__init__.py,sha256=4mPVzme2_-M64bgBu1cANIfBFAL0voa5I71-ceMr_qk,64
 homa/rl/DQN.py,sha256=PaNq9Z1K87IQ7Y7mhiJ1CE4TofgV7c7m1py8qT09vE4,20
 homa/rl/DRQN.py,sha256=zooojji9aeeubOP7cRPSHg31u2Assxk-qjXyGUWIO3A,49
 homa/rl/DiversityIsAllYouNeed.py,sha256=8yKzlVdLisForGyXqxaXUAWG_dozq7dNY8MBasCvniE,3322
-homa/rl/SoftActorCritic.py,sha256=N8EsiYbsLH-dpT2EmqdYFG9KvHNfO3JX8SG2LPTy94s,1962
+homa/rl/SoftActorCritic.py,sha256=3XmEKZ5IEM5Z1-OIpOsGSHZgL_veuYEv6dM_JBCEwwE,2060
 homa/rl/__init__.py,sha256=EaNDkIzLH1Oy0Wc0aAyyVs4HVMcZS1tdHDh631LKSXs,146
-homa/rl/utils.py,sha256=ySNGwWFBgN3Phg7Xn99CIJVHRWw3AmU9KcY-GP4ZQBc,236
+homa/rl/utils.py,sha256=IqbN5aDLwovocpPbxgywuetjz7GQwh9aJ4WFIOtLP3g,232
 homa/rl/buffers/Buffer.py,sha256=YCESh9tFxgWOLzGQj_IA0zLJoZWDmz6gCNu1iYsGp1s,388
 homa/rl/buffers/DiversityIsAllYouNeedBuffer.py,sha256=Nwcqs3Q10x6OKZ-zWug4IcBc6RR1TwEIybuFQOtmftA,1612
 homa/rl/buffers/ImageBuffer.py,sha256=HSmMt82hmkL3ooBYo7c6YUtTsMz9TAA8CvPh3y8z3yg,65
-homa/rl/buffers/SoftActorCriticBuffer.py,sha256=JQ9Y6KeeQS5naO_JPONiks-HYXw7hiZZAbqpoWDZlNI,1797
+homa/rl/buffers/SoftActorCriticBuffer.py,sha256=7ZqxG-zwB1U4_id1ad2y86ih2seTjauT5v1oK7wG28I,1977
 homa/rl/buffers/__init__.py,sha256=h1AkCHs6isXbNtxpaZfLp6YudHj1KlnOvURE64vhRa4,190
 homa/rl/buffers/concerns/HasRecordAlternatives.py,sha256=D5aVlPZlnGm0GyGtikKb4wZqyO6zpyqR1IOETmAgLx4,362
 homa/rl/buffers/concerns/ResetsCollection.py,sha256=bZ8q4czYXo1jMtVCnnlG69OgiJ0AqSGY6CiKzJC6xtQ,215
@@ -118,8 +119,8 @@ homa/rl/diayn/modules/ContinuousActorModule.py,sha256=yeC117I5gkXZSidQhjwakjiY7G
 homa/rl/diayn/modules/CriticModule.py,sha256=OUenwCG0dG4PnK7Iq-jy7oCTv_Cn9s7bXRpro6Pvb40,956
 homa/rl/diayn/modules/DiscriminatorModule.py,sha256=D58dKBv4f6gtrpqMKLK8XAZpiMqKfS4sG6s3QcF8iGE,891
 homa/rl/diayn/modules/__init__.py,sha256=1Pgjr4FT5WG-AMh26NPEfbf5pK6I02B1x8HYsgyUCJ4,149
-homa/rl/sac/SoftActor.py,sha256=CxR58IFrZ6xlmBj_gq_abZfgdzlVD71c6wA6wQiVL2c,2142
-homa/rl/sac/SoftCritic.py,sha256=EOX1vpH7YVwDuy-RdFgEpIKGioE7si0awoAYrHMTv4g,2923
+homa/rl/sac/SoftActor.py,sha256=UfmWDMnMkItta04ERQbs-2XA6DoeTG6gXMZ34mfYgB0,2170
+homa/rl/sac/SoftCritic.py,sha256=WA5o24wkA16rCCVEZ3BRVhmzT5ELGtWlO97adrWaFD8,2932
 homa/rl/sac/__init__.py,sha256=8EIkOcVvxN94gGzcZoX2XTnvTsHqW6yBaZ2RdFwIveM,68
 homa/rl/sac/modules/DualSoftCriticModule.py,sha256=Ax28i7U-KnP4QJig-AeeCfpPYNvTT3DfvRMJI-f-TGY,749
 homa/rl/sac/modules/SoftActorModule.py,sha256=LQ4z7s8mE3wwb1JgxPs0QvnriZULK3_ULdhkt60Ffpw,1152
@@ -143,8 +144,8 @@ homa/vision/concerns/__init__.py,sha256=mrw1YvN-GpQPvMwDF00KxnFkksPKo23RWM4KRioU
 homa/vision/modules/ResnetModule.py,sha256=eFudBnILD6OmgQtcW_CQQ8aZ62NEa4HyZ15-lobTtt0,712
 homa/vision/modules/SwinModule.py,sha256=3ZtUcfyJt0NMGmIlGpN35MIJG9QsgcLdFniZH7NxZQo,1227
 homa/vision/modules/__init__.py,sha256=zVMYB9IAO_xZylC1-N3p8ymHgEkAE2sBbuVz8K5Y1kk,74
-homa-0.3.13.dist-info/METADATA,sha256=1oENQJVWWN-hpkF440JOK0IiOI-i-Vbx9zcl7y5_pcs,1760
-homa-0.3.13.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-homa-0.3.13.dist-info/entry_points.txt,sha256=tJZzjs-f2QvFe3ES8Qta8IE5sAbeE8-cyZ_UtbgqG4s,51
-homa-0.3.13.dist-info/top_level.txt,sha256=tmOfy2tuaAwc3W5-i6j61_vYJsXgR4ivBWkhJ3ZtJDc,5
-homa-0.3.13.dist-info/RECORD,,
+homa-0.3.19.dist-info/METADATA,sha256=svvWLh14x71sFlxH_ApH0yaQ6QdmayDByYiHZUrPrSM,1760
+homa-0.3.19.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+homa-0.3.19.dist-info/entry_points.txt,sha256=tJZzjs-f2QvFe3ES8Qta8IE5sAbeE8-cyZ_UtbgqG4s,51
+homa-0.3.19.dist-info/top_level.txt,sha256=tmOfy2tuaAwc3W5-i6j61_vYJsXgR4ivBWkhJ3ZtJDc,5
+homa-0.3.19.dist-info/RECORD,,