Update training_multiSample.py

mNSF/training_multiSample.py

@@ -114,76 +114,153 @@ def truncate_history(loss_history, epoch):
 
 # Class to check convergence of the model
 class ConvergenceChecker(object):
- def __init__(self,span,dtp="float64"):
- """
- Initialize the ConvergenceChecker with polynomial basis functions.
+  def __init__(self, span, dtp="float64"):
+  """
+  Initialize the ConvergenceChecker with polynomial basis functions.
  
- Args:
- span: Number of recent observations to consider
- dtp: Data type for computations
- """
- x = np.arange(span,dtype=dtp)
- x-= x.mean()
- X = np.column_stack((np.ones(shape=x.shape),x,x**2,x**3))
- self.U = np.linalg.svd(X,full_matrices=False)[0]
-
- def smooth(self,y): 
- """Apply smoothing to the input vector."""
- return self.U@(self.U.T@y)
-
- def subset(self,y,idx=-1): 
- """Extract a subset of the input vector."""
- span = self.U.shape[0]
- lo = idx-span+1
- if idx==-1:
- return y[lo:]
- else:
- return y[lo:(idx+1)]
-
- def relative_change(self,y,idx=-1,smooth=True):
- """Calculate the relative change in the input vector."""
- y = self.subset(y,idx=idx)
- if smooth:
- y = self.smooth(y)
- prev=y[-2]
- return (y[-1]-prev)/(0.1+abs(prev))
-
-
-
-
- def converged(self,y,tol=1e-4,**kwargs):
- """Check if the relative change is below the tolerance."""
- return abs(self.relative_change(y,**kwargs)) < tol
-
-
- def relative_chg_normalized(self, y, idx_current = -1, len_trace=50, smooth=True):
- cc = self.relative_change_trace( y, idx_current = idx_current, len_trace=len_trace, smooth=smooth)
- print(cc)
- mean_cc = np.nanmean(cc) 
- sd_cc = np.std(cc) 
- return mean_cc/sd_cc
-
-
-
- def relative_change_trace(self, y, idx_current = -1, len_trace=50, smooth=True):
- #n = len(y)
- #span = self.U.shape[0]
- #span = self.U.shape[0]
- cc=self.relative_change_all(y,smooth=smooth)
- return cc[(idx_current-len_trace):idx_current]
-
-
- def relative_change_all(self,y,smooth=True):
- n = len(y)
- span = self.U.shape[0]
- cc = np.tile([np.nan],n)
- for i in range(span,n):
- cc[i] = self.relative_change(y,idx=i,smooth=smooth)
- return cc
-
- def converged_all(self,y,tol=1e-4,smooth=True):
- cc = self.relative_change_all(y,smooth=smooth)
- return np.abs(cc)<tol
+ Args:
+ span: Number of recent observations to consider
+ dtp: Data type for computations
+ """
+ # Create an array of indices
+ x = np.arange(span, dtype=dtp)
+ # Center the indices
+ x -= x.mean()
+ # Create a matrix of polynomial basis functions (constant, linear, quadratic, cubic)
+ X = np.column_stack((np.ones(shape=x.shape), x, x**2, x**3))
+ # Compute the left singular vectors of X
+ self.U = np.linalg.svd(X, full_matrices=False)[0]
+
+ def smooth(self, y):
+ """
+ Apply smoothing to the input vector using the precomputed basis.
+ 
+ Args:
+ y: Input vector to smooth
+ 
+ Returns:
+ Smoothed version of y
+ """
+ return self.U @ (self.U.T @ y)
+
+ def subset(self, y, idx=-1):
+ """
+ Extract a subset of the input vector.
+ 
+ Args:
+ y: Input vector
+ idx: Index to end the subset (default: -1, meaning the end of the vector)
+ 
+ Returns:
+ Subset of y
+ """
+ span = self.U.shape[0]
+ lo = idx - span + 1
+ if idx == -1:
+ return y[lo:]
+ else:
+ return y[lo:(idx+1)]
+
+ def relative_change(self, y, idx=-1, smooth=True):
+ """
+ Calculate the relative change in the input vector.
+ 
+ Args:
+ y: Input vector
+ idx: Index to calculate change at (default: -1)
+ smooth: Whether to apply smoothing (default: True)
+ 
+ Returns:
+ Relative change in y
+ """
+ y = self.subset(y, idx=idx)
+ if smooth:
+ y = self.smooth(y)
+ prev = y[-2]
+ # Calculate relative change with a small constant (0.1) to avoid division by zero
+ return (y[-1] - prev) / (0.1 + abs(prev))
+
+ def converged(self, y, tol=1e-4, **kwargs):
+ """
+ Check if the relative change is below the tolerance.
+ 
+ Args:
+ y: Input vector
+ tol: Convergence tolerance (default: 1e-4)
+ **kwargs: Additional arguments passed to relative_change
+ 
+ Returns:
+ Boolean indicating whether convergence criterion is met
+ """
+ return abs(self.relative_change(y, **kwargs)) < tol
+
+ def relative_chg_normalized(self, y, idx_current=-1, len_trace=50, smooth=True):
+ """
+ Calculate the normalized relative change over a trace of values.
+ 
+ Args:
+ y: Input vector
+ idx_current: Current index (default: -1)
+ len_trace: Length of the trace to consider (default: 50)
+ smooth: Whether to apply smoothing (default: True)
+ 
+ Returns:
+ Normalized relative change
+ """
+ cc = self.relative_change_trace(y, idx_current=idx_current, len_trace=len_trace, smooth=smooth)
+ print(cc)
+ mean_cc = np.nanmean(cc) 
+ sd_cc = np.std(cc) 
+ return mean_cc / sd_cc
+
+ def relative_change_trace(self, y, idx_current=-1, len_trace=50, smooth=True):
+ """
+ Calculate the trace of relative changes.
+ 
+ Args:
+ y: Input vector
+ idx_current: Current index (default: -1)
+ len_trace: Length of the trace to consider (default: 50)
+ smooth: Whether to apply smoothing (default: True)
+ 
+ Returns:
+ Array of relative changes
+ """
+ cc = self.relative_change_all(y, smooth=smooth)
+ return cc[(idx_current - len_trace):idx_current]
+
+ def relative_change_all(self, y, smooth=True):
+ """
+ Calculate relative changes for all possible subsets of y.
+ 
+ Args:
+ y: Input vector
+ smooth: Whether to apply smoothing (default: True)
+ 
+ Returns:
+ Array of relative changes for all subsets
+ """
+ n = len(y)
+ span = self.U.shape[0]
+ cc = np.tile([np.nan], n)
+ for i in range(span, n):
+ cc[i] = self.relative_change(y, idx=i, smooth=smooth)
+ return cc
+
+ def converged_all(self, y, tol=1e-4, smooth=True):
+ """
+ Check convergence for all possible subsets of y.
+ 
+ Args:
+ y: Input vector
+ tol: Convergence tolerance (default: 1e-4)
+ smooth: Whether to apply smoothing (default: True)
+ 
+ Returns:
+ Boolean array indicating convergence for each subset
+ """
+ cc = self.relative_change_all(y, smooth=smooth)
+ return np.abs(cc) < tol
 
 class ModelTrainer(object): #goal to change this to tf.module?
  """