better logging of checkpoints

Author: uym2

Diff for: CHANGELOG.md

@@ -1,3 +1,5 @@
+* LAML version 0.0.5
+    * better checkpoint logging
 * LAML version 0.0.4
     * add functionalities to create character matrices from fastq files (borrowed from a module of Cassiopeia) 
     * handle mismatch between character matrix and prior pickle

Diff for: laml_libs/Topology_search.py

@@ -68,7 +68,7 @@ def __accept_proposal__(self,curr_score,new_score,t):
         p = min(exp((new_score-curr_score-1e-12)/T),1)
         return random() < p
 
-    def search(self,resolve_polytomies=True,maxiter=100,verbose=False,nreps=1,strategy=DEFAULT_STRATEGY,checkpoint_file="laml_topo_search._ckpt.txt"):
+    def search(self,resolve_polytomies=True,maxiter=100,verbose=False,nreps=1,strategy=DEFAULT_STRATEGY,checkpoint_file=None):
         original_topos = self.treeTopoList
         original_params = self.params
         #nni_replicates = [(None,None)]*nreps
@@ -132,7 +132,7 @@ def search(self,resolve_polytomies=True,maxiter=100,verbose=False,nreps=1,strate
 
         return best_trees,best_score,best_params
 
-    def __search_one__(self,strategy,maxiter=100,verbose=False,only_marked=False, checkpoint_file="laml_topo_search._ckpt.txt"):
+    def __search_one__(self,strategy,maxiter=100,verbose=False,only_marked=False, checkpoint_file=None):
         # optimize branch lengths and other parameters for the starting tree
         mySolver = self.get_solver()
         score_tree_strategy = deepcopy(strategy)
@@ -164,8 +164,8 @@ def __search_one__(self,strategy,maxiter=100,verbose=False,only_marked=False, ch
                 print("Current score: " + str(curr_score))
                 stop_time = timeit.default_timer()
                 print("Runtime (s):", stop_time - start_time)
-            if nni_iter % 50 == 0:
-                with open(checkpoint_file, "w") as fout:
+            if nni_iter % chkpt_freq == 0 and checkpoint_file is not None:
+                with open(checkpoint_file, "a") as fout:
                     fout.write(f"NNI Iteration: {nni_iter}\n")
                     fout.write(f"Current newick tree: {best_trees}\n")
                     fout.write(f"Current negative-llh: {best_score}\n")

Diff for: laml_libs/__init__.py

@@ -17,4 +17,5 @@
 nni_conv_eps = 1e-15 # additional NNI convergence threshold since conv_eps has several purposes
 dmin = 0.005
 dmax = 10
+chkpt_freq = 10
 DEFAULT_STRATEGY={'resolve_search_only':False,'only_marked':False,'ultra_constr':False,'fixed_phi':None,'fixed_nu':None,'local_brlen_opt':True}

Diff for: run_laml.py

@@ -212,8 +212,7 @@ def main():
                 print("Running topology search in parallel...")
             else:
                 print("Running topology search sequentially...")
-            randval = int(random.random() * 1000)
-            checkpoint_file = f"{prefix}._ckpt.{randval}.txt"
+            checkpoint_file = f"{prefix}_ckpt.txt"
             opt_trees,max_score,opt_params = myTopoSearch.search(resolve_polytomies=resolve_polytomies,maxiter=args["maxIters"], verbose=args["verbose"], strategy=my_strategy, nreps=args['randomreps'],checkpoint_file=checkpoint_file) 
             nllh = -max_score