Bump prefix-dev/setup-pixi from 0.10.1 to 0.11.0

.github/workflows/ci.yaml

@@ -26,7 +26,7 @@ jobs:
     - uses: actions/checkout@v7
 
     - name: Set up virtual environment
-      uses: prefix-dev/setup-pixi@v0.10.1
+      uses: prefix-dev/setup-pixi@v0.11.0
       with:
         environments: ${{ matrix.environment }}
         frozen: true

.pre-commit-config.yaml

@@ -7,11 +7,11 @@ repos:
   - id: check-yaml
   - id: debug-statements
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.15.13
+  rev: v0.17.0
   hooks:
     - id: ruff-check
     - id: ruff-format
 - repo: https://github.com/tox-dev/pyproject-fmt
-  rev: v2.21.0
+  rev: v2.30.3
   hooks:
     - id: pyproject-fmt

docs/installation.md

@@ -72,6 +72,7 @@ To verify that DESCENT is installed correctly, run:
 
 ```python
 import descent
+
 print(descent.__version__)
 ```
 

docs/optimization.md

@@ -22,11 +22,7 @@ Adaptive Moment Estimation (Adam) is the default optimizer:
 ```python
 from descent.optim import Adam
 
-optimizer = Adam(
-    lr=1e-3,
-    betas=(0.9, 0.999),
-    eps=1e-8
-)
+optimizer = Adam(lr=1e-3, betas=(0.9, 0.999), eps=1e-8)
 ```
 
 **Pros:**
@@ -48,7 +44,7 @@ from descent.optim import AdamW
 optimizer = AdamW(
     lr=1e-3,
     weight_decay=1e-4,  # L2 regularization
-    betas=(0.9, 0.999)
+    betas=(0.9, 0.999),
 )
 ```
 
@@ -61,11 +57,7 @@ Stochastic Gradient Descent with momentum:
 ```python
 from descent.optim import SGD
 
-optimizer = SGD(
-    lr=1e-2,
-    momentum=0.9,
-    nesterov=True
-)
+optimizer = SGD(lr=1e-2, momentum=0.9, nesterov=True)
 ```
 
 **Pros:**
@@ -83,11 +75,7 @@ Limited-memory BFGS for quasi-Newton optimization:
 ```python
 from descent.optim import LBFGS
 
-optimizer = LBFGS(
-    lr=1.0,
-    max_iter=20,
-    history_size=10
-)
+optimizer = LBFGS(lr=1.0, max_iter=20, history_size=10)
 ```
 
 **Pros:**
@@ -116,11 +104,7 @@ Reduce learning rate at specified epochs:
 ```python
 from torch.optim.lr_scheduler import StepLR
 
-scheduler = StepLR(
-    optimizer,
-    step_size=30,
-    gamma=0.1
-)
+scheduler = StepLR(optimizer, step_size=30, gamma=0.1)
 ```
 
 ### Reduce on Plateau
@@ -131,11 +115,7 @@ Reduce when loss plateaus:
 from torch.optim.lr_scheduler import ReduceLROnPlateau
 
 scheduler = ReduceLROnPlateau(
-    optimizer,
-    mode='min',
-    factor=0.5,
-    patience=10,
-    threshold=1e-4
+    optimizer, mode="min", factor=0.5, patience=10, threshold=1e-4
 )
 ```
 
@@ -146,11 +126,7 @@ Cosine decay with warm restarts:
 ```python
 from torch.optim.lr_scheduler import CosineAnnealingWarmRestarts
 
-scheduler = CosineAnnealingWarmRestarts(
-    optimizer,
-    T_0=10,
-    T_mult=2
-)
+scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=10, T_mult=2)
 ```
 
 ### Exponential Decay
@@ -160,10 +136,7 @@ Exponential learning rate decay:
 ```python
 from torch.optim.lr_scheduler import ExponentialLR
 
-scheduler = ExponentialLR(
-    optimizer,
-    gamma=0.95
-)
+scheduler = ExponentialLR(optimizer, gamma=0.95)
 ```
 
 ## Gradient Computation
@@ -211,11 +184,7 @@ For problems with structured residuals, DESCENT provides Levenberg-Marquardt opt
 ```python
 from descent.optim import LevenbergMarquardt
 
-optimizer = LevenbergMarquardt(
-    lambda_init=1e-3,
-    lambda_factor=10.0,
-    max_iter=100
-)
+optimizer = LevenbergMarquardt(lambda_init=1e-3, lambda_factor=10.0, max_iter=100)
 ```
 
 ## Regularization
@@ -229,7 +198,7 @@ Add L2 penalty to loss:
 optimizer = AdamW(lr=1e-3, weight_decay=1e-4)
 
 # Or manually
-loss = target_loss + 1e-4 * torch.sum(parameters ** 2)
+loss = target_loss + 1e-4 * torch.sum(parameters**2)
 ```
 
 ### L1 Regularization
@@ -301,13 +270,13 @@ from itertools import product
 lr_values = [1e-2, 1e-3, 1e-4]
 wd_values = [0, 1e-4, 1e-3]
 
-best_loss = float('inf')
+best_loss = float("inf")
 best_params = None
 
 for lr, wd in product(lr_values, wd_values):
     optimizer = AdamW(lr=lr, weight_decay=wd)
     loss = train(optimizer, ...)
-    
+
     if loss < best_loss:
         best_loss = loss
         best_params = (lr, wd)

docs/targets.md

@@ -23,7 +23,7 @@ target = EnergyTarget(
     dataset="qm_energies.json",
     weight=1.0,
     denominator="std",  # Normalize by standard deviation
-    energy_type="absolute"  # or "relative"
+    energy_type="absolute",  # or "relative"
 )
 ```
 
@@ -69,7 +69,7 @@ target = ThermoTarget(
     weight=0.5,
     properties=["density", "hvap", "dielectric"],
     temperatures=[298.15],  # K
-    pressures=[1.0]  # atm
+    pressures=[1.0],  # atm
 )
 ```
 
@@ -150,41 +150,42 @@ Create custom target functions by subclassing `BaseTarget`:
 from descent.targets import BaseTarget
 import torch
 
+
 class CustomTarget(BaseTarget):
     def __init__(self, dataset, weight=1.0, **kwargs):
         super().__init__(weight=weight)
         self.dataset = self.load_dataset(dataset)
-        
+
     def compute_loss(self, system, parameters):
         """
         Compute loss for this target.
-        
+
         Parameters
         ----------
         system : System
             The molecular system
         parameters : dict
             Current force field parameters
-            
+
         Returns
         -------
         loss : torch.Tensor
             Scalar loss value
         """
         predictions = self.predict(system, parameters)
         references = self.get_references()
-        
+
         loss = torch.mean((predictions - references) ** 2)
         return loss
-        
+
     def predict(self, system, parameters):
         """Compute predictions using current parameters."""
         # Your implementation here
         pass
-        
+
     def get_references(self):
         """Get reference values."""
-        return torch.tensor(self.dataset['values'])
+        return torch.tensor(self.dataset["values"])
 ```
 
 ## Combining Targets
@@ -200,7 +201,7 @@ config = TrainingConfig(
         {"type": "energy", "dataset": "qm.json", "weight": 1.0},
         {"type": "thermo", "dataset": "exp.json", "weight": 0.5},
         {"type": "dimer", "dataset": "dimers.json", "weight": 0.3},
-    ]
+    ],
 )
 ```
 

docs/training.md

@@ -29,7 +29,7 @@ config = TrainingConfig(
     force_field="openff-2.0.0.offxml",
     parameters=["[#6:1]-[#6:2]", "[#6:1]-[#1:2]"],
     targets=[...],
-    optimizer_config={...}
+    optimizer_config={...},
 )
 
 # Run training
@@ -50,7 +50,7 @@ from descent.targets import EnergyTarget
 target = EnergyTarget(
     dataset="energies.json",
     weight=1.0,
-    denominator="std"  # Normalize by standard deviation
+    denominator="std",  # Normalize by standard deviation
 )
 ```
 
@@ -61,11 +61,7 @@ Fit to experimental thermodynamic properties (density, enthalpy of vaporization,
 ```python
 from descent.targets import ThermoTarget
 
-target = ThermoTarget(
-    dataset="thermo.json",
-    weight=0.5,
-    properties=["density", "hvap"]
-)
+target = ThermoTarget(dataset="thermo.json", weight=0.5, properties=["density", "hvap"])
 ```
 
 ### Dimer Targets
@@ -75,10 +71,7 @@ Fit to dimer interaction energies:
 ```python
 from descent.targets import DimerTarget
 
-target = DimerTarget(
-    dataset="dimers.json",
-    weight=0.3
-)
+target = DimerTarget(dataset="dimers.json", weight=0.3)
 ```
 
 ## Optimization Strategies
@@ -88,10 +81,7 @@ DESCENT uses PyTorch optimizers for parameter optimization:
 ```python
 from descent.optim import AdamW
 
-optimizer = AdamW(
-    lr=1e-3,
-    weight_decay=1e-4
-)
+optimizer = AdamW(lr=1e-3, weight_decay=1e-4)
 ```
 
 Common optimizers:
@@ -107,11 +97,9 @@ The overall loss is a weighted combination of target losses:
 ```python
 from descent.utils.loss import WeightedLoss
 
-loss_fn = WeightedLoss(targets=[
-    (energy_target, 1.0),
-    (thermo_target, 0.5),
-    (dimer_target, 0.3)
-])
+loss_fn = WeightedLoss(
+    targets=[(energy_target, 1.0), (thermo_target, 0.5), (dimer_target, 0.3)]
+)
 ```
 
 ## Training Loop
@@ -125,12 +113,7 @@ from descent.train import train
 config = TrainingConfig(...)
 
 # Run training
-results = train(
-    config,
-    max_epochs=100,
-    batch_size=32,
-    validation_split=0.2
-)
+results = train(config, max_epochs=100, batch_size=32, validation_split=0.2)
 
 # Access results
 print(f"Final loss: {results['final_loss']}")
@@ -158,17 +141,17 @@ Save and resume training:
 ```python
 # Save checkpoint
 checkpoint = {
-    'epoch': epoch,
-    'parameters': parameters,
-    'optimizer_state': optimizer.state_dict(),
-    'loss': loss
+    "epoch": epoch,
+    "parameters": parameters,
+    "optimizer_state": optimizer.state_dict(),
+    "loss": loss,
 }
-torch.save(checkpoint, 'checkpoint.pt')
+torch.save(checkpoint, "checkpoint.pt")
 
 # Resume training
-checkpoint = torch.load('checkpoint.pt')
-parameters = checkpoint['parameters']
-optimizer.load_state_dict(checkpoint['optimizer_state'])
+checkpoint = torch.load("checkpoint.pt")
+parameters = checkpoint["parameters"]
+optimizer.load_state_dict(checkpoint["optimizer_state"])
 ```
 
 ## Best Practices
@@ -189,6 +172,7 @@ Create custom target functions by subclassing `BaseTarget`:
 ```python
 from descent.targets import BaseTarget
 
+
 class CustomTarget(BaseTarget):
     def compute_loss(self, system, parameters):
         # Custom loss computation
@@ -203,12 +187,7 @@ Use PyTorch learning rate schedulers:
 ```python
 from torch.optim.lr_scheduler import ReduceLROnPlateau
 
-scheduler = ReduceLROnPlateau(
-    optimizer,
-    mode='min',
-    factor=0.5,
-    patience=10
-)
+scheduler = ReduceLROnPlateau(optimizer, mode="min", factor=0.5, patience=10)
 ```
 
 ### Regularization