moospread 0.1.3__py3-none-any.whl → 0.1.5__py3-none-any.whl

moospread/tasks/bo_torch.py

@@ -35,13 +35,17 @@ class BraninCurrin(PymooProblemTorch):
 
     """
     
-    def __init__(self, path = None, ref_point=None, negate=True, **kwargs):
+    def __init__(self, path = None, ref_point=None, negate=False, **kwargs):
         super().__init__(n_var=2, n_obj=2, 
                          xl=torch.zeros(2, dtype=torch.float) + 1e-6,
                          xu=torch.ones(2, dtype=torch.float) - 1e-6,
                          vtype=float, **kwargs)
+        
+        self.negate = negate
         if ref_point is None:
             self.ref_point = [18.0, 6.0]
+            if self.negate:
+                self.ref_point = [-val for val in self.ref_point]
         else:
             self.ref_point = ref_point
             
@@ -50,7 +54,7 @@ class BraninCurrin(PymooProblemTorch):
         
     def _calc_pareto_front(self, n_pareto_points: int = 100) -> torch.Tensor:
         if self.path is not None:
-            front = np.loadtxt(self.path)
+            front = np.load(self.path)
             return torch.from_numpy(front).to(self.device)
         else:
             return None
@@ -85,6 +89,8 @@ class BraninCurrin(PymooProblemTorch):
         branin = self._rescaled_branin(X=X)
         currin = self._currin(X=X)
         out["F"] = torch.stack([branin, currin], dim=-1)
+        if self.negate:
+            out["F"] = -out["F"]
     
 class Penicillin(PymooProblemTorch):
     r"""A penicillin production simulator from [Liang2021]_.
@@ -125,13 +131,16 @@ class Penicillin(PymooProblemTorch):
     R = 1.9872  # CAL/(MOL K)
     V_max = 180.0
     
-    def __init__(self, path = None, ref_point=None, negate=True, **kwargs):
+    def __init__(self, path = None, ref_point=None, negate=False, **kwargs):
         super().__init__(n_var=7, n_obj=3, 
                          xl=torch.tensor([60.0, 0.05, 293.0, 0.05, 0.01, 500.0, 5.0], dtype=torch.float) + 1e-6,
                          xu=torch.tensor([120.0, 18.0, 303.0, 18.0, 0.5, 700.0, 6.5], dtype=torch.float) - 1e-6,
                          vtype=float, **kwargs)
+        self.negate = negate
         if ref_point is None:
             self.ref_point = [25.935, 57.612, 935.5]
+            if self.negate:
+                self.ref_point = [-val for val in self.ref_point]
         else:
             self.ref_point = ref_point
             
@@ -140,7 +149,7 @@ class Penicillin(PymooProblemTorch):
         
     def _calc_pareto_front(self, n_pareto_points: int = 100) -> torch.Tensor:
         if self.path is not None:
-            front = np.loadtxt(self.path)
+            front = np.load(self.path)
             return torch.from_numpy(front).to(self.device)
         else:
             return None
@@ -222,10 +231,11 @@ class Penicillin(PymooProblemTorch):
     def _evaluate(self, X: torch.Tensor, out: dict, *args, **kwargs) -> None:
         # This uses in-place operations. Hence, the clone is to avoid modifying
         # the original X in-place.
-        out["F"] = self.penicillin_vectorized(X.view(-1, self.dim).clone()).view(
-            *X.shape[:-1], self.num_objectives
+        out["F"] = self.penicillin_vectorized(X.view(-1, self.n_var).clone()).view(
+            *X.shape[:-1], self.n_obj
         )
-        
+        if self.negate:
+            out["F"] = -out["F"]
 
 class VehicleSafety(PymooProblemTorch):
     r"""Optimize Vehicle crash-worthiness.
@@ -239,13 +249,16 @@ class VehicleSafety(PymooProblemTorch):
     pareto front from [Tanabe2020]_.
     """
     
-    def __init__(self, path = None, ref_point=None, negate=True, **kwargs):
+    def __init__(self, path = None, ref_point=None, negate=False, **kwargs):
         super().__init__(n_var=5, n_obj=3, 
                          xl=torch.ones(5, dtype=torch.float) + 1e-6,
                          xu=3*torch.ones(5, dtype=torch.float) - 1e-6,
                          vtype=float, **kwargs)
+        self.negate = negate
         if ref_point is None:
             self.ref_point = [1864.72022, 11.81993945, 0.2903999384]
+            if self.negate:
+                self.ref_point = [-val for val in self.ref_point]
         else:
             self.ref_point = ref_point
             
@@ -254,7 +267,7 @@ class VehicleSafety(PymooProblemTorch):
         
     def _calc_pareto_front(self, n_pareto_points: int = 100) -> torch.Tensor:
         if self.path is not None:
-            front = np.loadtxt(self.path)
+            front = np.load(self.path)
             return torch.from_numpy(front).to(self.device)
         else:
             return None
@@ -298,3 +311,5 @@ class VehicleSafety(PymooProblemTorch):
         )
         f_X = torch.cat([f1, f2, f3], dim=-1)
         out["F"] = f_X
+        if self.negate:
+            out["F"] = -out["F"]

moospread/utils/mobo_utils/learning/model_init.py

@@ -4,7 +4,6 @@ from moospread.utils.mobo_utils.learning.model import NeuralNet
 import torch.nn as nn
 
 
-
 def init_dom_nn_classifier(x, y, rel_map, dom, n_var, batch_size=32,
                            activation='relu', lr=0.001, weight_decay=0.00001):
     device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
@@ -13,10 +12,7 @@ def init_dom_nn_classifier(x, y, rel_map, dom, n_var, batch_size=32,
     num_hidden_layers = 3
     epochs = 20
     data = prepare_dom_data(x, y, rel_map, dom, data_kind='tensor', device=device)
-    # print(f"Data shape: {data.shape}")
-    # print("data[:, -1]", data[:, -1])
     weight = compute_class_weight(data[:, -1])
-    # print(f"Class weights: {weight}")
     if weight is None:
         return None
 

moospread/utils/mobo_utils/learning/utils.py

@@ -4,8 +4,6 @@ import numpy as np
 import torch.nn as nn
 
 
-# This file implements some Utility Functions about training
-
 def prepare_dom_data(x, y, rel_map, dom, start=0, data_kind='tensor', device='cpu'):
     n = len(x)
     data = []
@@ -65,22 +63,48 @@ def load_batched_dom_data(data, batch_size):
 
 
 def compute_class_weight(class_labels):
+    #################################### Original implementation (can return None too eagerly):
+    # n_examples = len(class_labels)
+
+    # class_labels = class_labels.cpu().numpy()
+    # n_zero = np.sum(class_labels == 0)
+    # n_one = np.sum(class_labels == 1)
+    # n_two = n_examples - n_zero - n_one
+
+    # if n_zero == 0 or n_one == 0 or n_two == 0:
+    #     return None
+
+    # w_zero = n_examples / (3. * n_zero)
+    # w_one = n_examples / (3. * n_one)
+    # w_two = n_examples / (3. * n_two)
+    
+    # return w_zero, w_one, w_two
+
+    # return w_zero, w_one, w_two
+    ################################### Revised implementation:
+    eps = 1e-6
     n_examples = len(class_labels)
-
-    class_labels = class_labels.cpu().numpy()
-    n_zero = np.sum(class_labels == 0)
-    n_one = np.sum(class_labels == 1)
-    n_two = n_examples - n_zero - n_one
-
-    if n_zero == 0 or n_one == 0 or n_two == 0:
-        return None
-
-    w_zero = n_examples / (3. * n_zero)
-    w_one = n_examples / (3. * n_one)
-    w_two = n_examples / (3. * n_two)
-
-    return w_zero, w_one, w_two
-
+    labels = class_labels.detach().cpu().numpy().astype(int)
+    n_zero = np.sum(labels == 0)
+    n_one  = np.sum(labels == 1)
+    n_two  = np.sum(labels == 2)
+    counts = np.array([n_zero, n_one, n_two], dtype=float)
+    weights = np.zeros(3, dtype=float)
+    # indices of present classes
+    present = counts > 0
+    missing = counts == 0
+    if missing.any():
+        # set missing class(es) to eps
+        weights[missing] = eps
+        # compute raw weights for present classes using your formula
+        raw = n_examples / (3.0 * counts[present])
+        # normalize them to sum to (1 - eps * num_missing)
+        remaining = 1.0 - eps * missing.sum()
+        weights[present] = raw / raw.sum() * remaining
+    else:
+        # standard case: all classes present
+        weights = n_examples / (3.0 * counts)
+    return tuple(weights)
 
 def train_nn(data, data_loader, net, criterion, optimizer, batch_size, epochs):
     #10000 12

{moospread-0.1.3.dist-info → moospread-0.1.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: moospread
-Version: 0.1.3
+Version: 0.1.5
 Summary: Sampling-based Pareto front Refinement via Efficient Adaptive Diffusion
 Author-email: Sedjro Salomon Hotegni <salomon.hotegni@aims.ac.rw>
 Maintainer-email: Sedjro Salomon Hotegni <salomon.hotegni@tu-dortmund.de>
@@ -63,6 +63,20 @@ Dynamic: license-file
 </p>
 -->
 
+<p align="center">
+  <a href="https://pypi.org/project/moospread/"><img src="https://img.shields.io/pypi/v/moospread.svg" alt="PyPI version"></a>
+  <a href="https://moospread.readthedocs.io">
+  <img src="https://img.shields.io/badge/docs-online-brightgreen.svg" alt="Documentation">
+</a>
+</p>
+<div align="center">
+    <h3>
+      <a href="https://pypi.org/project/moospread/">Installation</a> |
+      <a href="https://moospread.readthedocs.io/en/latest/">Documentation</a> | 
+      <a href="https://openreview.net/forum?id=4731mIqv89">Paper</a>
+    </h3>
+</div>
+
 # SPREAD: Sampling-based Pareto front Refinement via Efficient Adaptive Diffusion
 
 > SPREAD is a novel sampling-based approach for multi-objective optimization that leverages diffusion models to efficiently refine and generate well-spread Pareto front approximations. It combines the expressiveness of diffusion models with multi-objective optimization principles to achieve both high convergence to the Pareto front and excellent diversity across the objective space. SPREAD demonstrates competitive performance against state-of-the-art methods while providing a flexible framework for different optimization contexts.
@@ -128,11 +142,10 @@ This will train a diffusion-based multi-objective solver, approximate the Pareto
 
 ---
 
-<!--
+
 ### 📚 Next steps
 
-For more advanced examples (offline mode, Bayesian mode, custom problems), see the full [documentation](https://moospread.readthedocs.io/en/latest/).
--->
+For more advanced examples (offline mode, mobo mode, tutorials), see the full [documentation](https://moospread.readthedocs.io/en/latest/).
 
 ## Citation
 If you find `moospread` useful in your research, please consider citing:

{moospread-0.1.3.dist-info → moospread-0.1.5.dist-info}/RECORD

@@ -1,8 +1,8 @@
 moospread/__init__.py,sha256=v9TLUZq0-q0j_23NB7S4ugJqogOMutUuL9MMCi4zu4I,124
-moospread/core.py,sha256=bCGx7XL14GurdNTyeH6trWs-mJsBrdeWQNQV1pSxOTs,91648
+moospread/core.py,sha256=U29SSbF6RLkncfZVlgrKGOlI3aXiSjPpN-bvu9Mb5T4,117510
 moospread/problem.py,sha256=YjT4k_K7qTZDhWzIYkaBQlsZRVnpP6iV3F8ShFhGAck,6042
 moospread/tasks/__init__.py,sha256=RrtuZs1TyL35fDzGjlm1PnFqxXd75Qdl-q3Sdld7dsw,309
-moospread/tasks/bo_torch.py,sha256=0I77aWBM4TzfH7l2iQLTRm_5Orc25rmXabrcpPQpVb8,10371
+moospread/tasks/bo_torch.py,sha256=a5g47Mckkj4Ci-VbchAl2PsNXZ6pWAiEvxMqETm8rbs,10898
 moospread/tasks/dtlz_torch.py,sha256=dhmzUj-dbhF3zXIOgx2Z-PpuIqvxAttnEoBM0C9nUt0,5134
 moospread/tasks/mw_torch.py,sha256=hvYGxcaCr-AFgZd_-rLcIul_a5cZrOV3dRVB5sl9Wuo,9585
 moospread/tasks/re_torch.py,sha256=TK7i-9i7qlDZC2H4vdhcVgHybarKwJ-M27yAibAg3LI,14062
@@ -26,10 +26,10 @@ moospread/utils/mobo_utils/evolution/norm.py,sha256=V94LKVTF3ymVdEsI9rMc-4lwXh1Z
 moospread/utils/mobo_utils/evolution/utils.py,sha256=bfrUxmFAnjTr6UwY-8AifLZvt2KMj_NR85neQHhwhlU,2314
 moospread/utils/mobo_utils/learning/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moospread/utils/mobo_utils/learning/model.py,sha256=9PS2BdBr70jOllm6snYbP2iUAbGcJWRAmX_EiQEwnYg,1372
-moospread/utils/mobo_utils/learning/model_init.py,sha256=s7eCdKcF-c9mHpjKnSBuovKwONwqu1m-KXanc5ej6ng,1278
+moospread/utils/mobo_utils/learning/model_init.py,sha256=oUSljcQYhPKJMd9YT9pqDd9dLAJ31GM8GAOPjGLr73A,1156
 moospread/utils/mobo_utils/learning/model_update.py,sha256=csx2k-7kaYbGaiNlFe_kwdaTWuErxcqjTo224ONiq_Q,1447
 moospread/utils/mobo_utils/learning/prediction.py,sha256=eQec9R3omJNC5oKecQWxdJ2Y7fbHeK3YfodkHwrtOZ0,3173
-moospread/utils/mobo_utils/learning/utils.py,sha256=fgwekoEfc5owsa5KmXNGmnlIgnlANwpV00xwUD_eRr8,3992
+moospread/utils/mobo_utils/learning/utils.py,sha256=jCigiQVA41jsMmDek5WMCcxYTkprxIL7TSezHZPWAJ4,5041
 moospread/utils/mobo_utils/mobo/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moospread/utils/mobo_utils/mobo/acquisition.py,sha256=83CP4TMG0A29KEgPiiR588e3FZfsPx2hVZCBqaW6c1U,7293
 moospread/utils/mobo_utils/mobo/algorithms.py,sha256=MX1-jhYivoRHi4BAtUJsmTd5gK5nLLkQv7hpJJzWHlw,1518
@@ -57,8 +57,8 @@ moospread/utils/mobo_utils/mobo/surrogate_model/thompson_sampling.py,sha256=Nmp6
 moospread/utils/offline_utils/__init__.py,sha256=MJC-fqvQnbQ0T_wjCw_QK8nKo_xpQxh0buq91fxYjFY,742
 moospread/utils/offline_utils/handle_task.py,sha256=kgHpWotCIrmDBjs08KXyp1BuwXe5APkw3q7cu-xWlz4,8365
 moospread/utils/offline_utils/proxies.py,sha256=DPBykB8l1XJmT5QQCAQrgMZz-8FiGEiNwN0bBdYJIaY,11218
-moospread-0.1.3.dist-info/licenses/LICENSE,sha256=YwtV5PRo6WMw5CWQMD728fSF8cWEKKfwOhek37Yi1so,1079
-moospread-0.1.3.dist-info/METADATA,sha256=epF8DlyXLJAlXpOx2h8FlyG5FByj0aV4p8EefoHLU6k,5645
-moospread-0.1.3.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-moospread-0.1.3.dist-info/top_level.txt,sha256=LWi5kIahDQRNXNkx55T-gefn09Bgcq8SoCxp72S-7x0,10
-moospread-0.1.3.dist-info/RECORD,,
+moospread-0.1.5.dist-info/licenses/LICENSE,sha256=YwtV5PRo6WMw5CWQMD728fSF8cWEKKfwOhek37Yi1so,1079
+moospread-0.1.5.dist-info/METADATA,sha256=WPouMOqtqRjr7RUggiJrzqDd-uN4my3EFJq8NtoxThk,6189
+moospread-0.1.5.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+moospread-0.1.5.dist-info/top_level.txt,sha256=LWi5kIahDQRNXNkx55T-gefn09Bgcq8SoCxp72S-7x0,10
+moospread-0.1.5.dist-info/RECORD,,