From 733a888299d2d0691e1c8d39c4522f15ea29c3ee Mon Sep 17 00:00:00 2001
From: Luigi Bonati <lui.bonati@gmail.com>
Date: Tue, 2 May 2023 10:30:22 +0200
Subject: [PATCH] removed reduced rank algorithm from main branch (see #54)

---
 mlcolvar/core/stats/tica.py | 61 ++++---------------------------------
 1 file changed, 6 insertions(+), 55 deletions(-)

diff --git a/mlcolvar/core/stats/tica.py b/mlcolvar/core/stats/tica.py
index ed66d6c5..094f5f91 100644
--- a/mlcolvar/core/stats/tica.py
+++ b/mlcolvar/core/stats/tica.py
@@ -41,7 +41,7 @@ def extra_repr(self) -> str:
         repr = f"in_features={self.in_features}, out_features={self.out_features}"
         return repr
 
-    def compute(self, data, weights = None, remove_average=True, save_params=False, algorithm = 'least_squares'):
+    def compute(self, data, weights = None, remove_average=True, save_params=False):
         """Perform TICA computation.
 
         Parameters
@@ -54,17 +54,12 @@ def compute(self, data, weights = None, remove_average=True, save_params=False,
             whether to make the inputs mean free, by default True
         save_params : bool, optional
             Save parameters of estimator, by default False
-        algorithm : str, optional
-            Algorithm to use, by default 'least_squares'. Options are 'least_squares' and 'reduced_rank'. Both algorithms are described in [1]_.
+
         Returns
         -------
-        torch.Tensor
-            Eigenvalues
+        torch.Tensor,torch.Tensor
+            eigenvalues,eigenvectors
         
-        References
-        ----------
-        .. [1] V. Kostic, P. Novelli, A. Maurer, C. Ciliberto, L. Rosasco, and M. Pontil, "Learning Dynamical Systems via Koopman Operator Regression in Reproducing Kernel Hilbert Spaces" (2022).
-
         """
         # parse args
 
@@ -81,16 +76,7 @@ def compute(self, data, weights = None, remove_average=True, save_params=False,
         C_0 = correlation_matrix(x_t,x_t,w_t)
         C_lag = correlation_matrix(x_t,x_lag,w_lag)
 
-        if (algorithm == 'reduced_rank') and (self.out_features >= self.in_features):
-            warnings.warn('out_features is greater or equal than in_features. reduced_rank is equal to least_squares.')
-            algorithm = 'least_squares'
-        
-        if algorithm == 'reduced_rank':
-            evals, evecs = reduced_rank_eig(C_0, C_lag, self.reg_C_0, rank = self.out_features)
-        elif algorithm != 'least_squares':
-            raise ValueError(f'algorithm {algorithm} not recognized. Options are least_squares and reduced_rank.')
-        else:
-            evals, evecs = cholesky_eigh(C_lag,C_0,self.reg_C_0,n_eig=self.out_features) 
+        evals, evecs = cholesky_eigh(C_lag,C_0,self.reg_C_0,n_eig=self.out_features) 
             
         if save_params:
             self.evals = evals
@@ -176,40 +162,5 @@ def test_tica():
     s2 = tica(X2)
     print(X2.shape,'-->',s2.shape)
 
-def test_reduced_rank_tica():
-    in_features = 10
-    X = torch.rand(100,in_features)*100
-    x_t = X[:-1]
-    x_lag = X[1:]
-    w_t = torch.rand(len(x_t))
-    w_lag = w_t
-    
-    # direct way, compute tica function
-    tica = TICA(in_features,out_features=5)
-    print(tica)
-    tica.compute([x_t,x_lag],[w_t,w_lag], save_params=True, algorithm='reduced_rank')
-    s = tica(X)
-    print(X.shape,'-->',s.shape)
-    print('eigvals',tica.evals)
-    print('timescales', tica.timescales(lag=10))
-
-    # step by step
-    tica = TICA(in_features)
-    C_0 = correlation_matrix(x_t,x_t)
-    C_lag = correlation_matrix(x_t,x_lag)
-    print(C_0.shape,C_lag.shape)
-
-    evals, evecs = reduced_rank_eig(C_0, C_lag, 1e-6, rank = 5)
-    print(evals.shape,evecs.shape)
-
-    print('>> batch') 
-    s = tica(X)
-    print(X.shape,'-->',s.shape)
-    print('>> single')
-    X2 = X[0]
-    s2 = tica(X2)
-    print(X2.shape,'-->',s2.shape)
-
 if __name__ == "__main__":
-    test_tica()
-    test_reduced_rank_tica()
\ No newline at end of file
+    test_tica()
\ No newline at end of file