Performance tests

csirmaz · May 14, 2023 · fe02166 · fe02166
1 parent 0040f91
commit fe02166
Show file tree

Hide file tree

Showing 2 changed files with 43 additions and 21 deletions.
diff --git a/csequence.py b/csequence.py
@@ -169,7  169,7 @@ def as_set(self):
 
 
     @classmethod
-    def from_set_union(cls, command_sets):
     def from_set_union(cls, command_sets, return_length=False):
         """Turn the union of some command sets into a command sequence; the order of commands is not guaranteed
 
         Arguments:
@@ -191,7  191,9 @@ def from_set_union(cls, command_sets):
             if prev_command is None or (not command.equals(prev_command)):
                 union.append(command)
             prev_command = command
-        print(f"Union is {len(union)} long for:")
 
         if return_length:
             return (CSequence(union), len(union))
         return CSequence(union)
 
 
@@ -411,7  413,7 @@ def set_bit(node, bit):
 
 
     @classmethod
-    def get_any_merger(cls, command_sets, decisions=None, debug=False):
     def get_any_merger(cls, command_sets, decisions=None, debug=False, return_lengths=False):
         """Given a set of jointly refluent canonical command sets, generate a merger.
         Suitable to produce all possible mergers; see Session.get_all_mergers().
 
@@ -512,8  514,12 @@ def mark_delete_destructors_up(command):
                     continue
                 break            
 
-        # from_set_union() applies order_by_node_value()        
-        union = cls.from_set_union(command_sets).add_up_pointers().add_backlinks()
         # from_set_union() applies order_by_node_value()
         if return_lengths:
             union, len_union = cls.from_set_union(command_sets, return_length=True)
         else:
             union = cls.from_set_union(command_sets)
         union = union.add_up_pointers().add_backlinks()
 
         # (0) Initialise flags in top-down order
         for command in union.forward():
@@ -703,5  709,6 @@ def process_node_commands_4(commands):
             for i, d in enumerate(decisions):
                 print(f"  #{i} {d['current_decision']} of {d['num_options']} for {d['comment']}")
 
-        print(f"  merger_size={len(merger)}")
         if return_lengths:
             return (decisions, CSequence(merger), {'union': len_union, 'merger': len(merger)})
         return (decisions, CSequence(merger))
diff --git a/test_on_set.py b/test_on_set.py
@@ -139,16  139,25 @@ def generate_sequences(self, num_users):
                 })
 
 
 num_experiments = 10
 
 # CSV header
 csv_hdr = ([
     'spread', 'size', 'num_users', 'num_mergers', 'max_nodes', 'num_nodes', 'seq_len', 'union_len', 'merger_len', 
     'avg_time', 'mse', 'merger_time', 'command_time'
     ]   [f"t{i}" for i in range(num_experiments)])
 print(",".join(csv_hdr))
 
 experiments = {}
-for experiment in range(10):
 for experiment in range(num_experiments):
     random.shuffle(settings)
-    for setting in settings:
     for set_ix, setting in enumerate(settings):
         spread = setting['spread']
         size = setting['size']
         num_users = setting['num_users']
         num_mergers = setting['num_mergers']
 
-        max_size = size*(2*spread 1)   size*(2*spread 1)*(2*spread 1)   size*(2*spread 1)*(2*spread 1)*size
         max_nodes = size*(2*spread 1)   size*(2*spread 1)*(2*spread 1)   size*(2*spread 1)*(2*spread 1)*size
 
         if False and experiment == 0:
             test1 = Test(size=size, spread=spread, num_users=num_users)
@@ -165,7  174,7 @@ def generate_sequences(self, num_users):
             test = Test(size=size, spread=spread, num_users=num_users) # reset flags, etc.
             gc.collect()
             start = timer()
-            decisions, merger = CSequence.get_any_merger(test.sequences, decisions=decisions, debug=False)
             decisions, merger, lengths = CSequence.get_any_merger(test.sequences, decisions=decisions, debug=False, return_lengths=True)
             end = timer()
             time_spent  = end - start
             if decisions is None: # no more mergers
@@ -175,15  184,21 @@ def generate_sequences(self, num_users):
                 break
 
         if i == num_mergers: # We have enough mergers
-            key = f"spread={spread} size={size} users={num_users} mergers={num_mergers} -> max_nodes={max_size} num_nodes={test.num_nodes} seq_len={test.sequence_length}"
-            print(f"{key} #{experiment} -> {time_spent}s")
             exp_data = [spread, size, num_users, num_mergers, max_nodes, test.num_nodes, test.sequence_length, lengths['union'], lengths['merger']]
             key = ":".join([str(x) for x in exp_data])
             if key not in experiments:
-                experiments[key] = []
-            experiments[key].append(time_spent)
-
-    time.sleep(30)
-
-print("----- AGGREGATING ------")
-for k, times in experiments.items():
-    avg = sum(times)/len(times)
-    print(f"{k} -> times: " " ".join([str(t) for t in times]) f" average: {avg} s")
                 experiments[key] = {'data': exp_data, 'times':[]}
             print(f"PROGRESS {(experiment*len(settings) set_ix)/(num_experiments*len(settings))*100}%")
             experiments[key]['times'].append(time_spent)
 
     if experiment < num_experiments - 1:
         time.sleep(30)
 
 # CSV lines
 import numpy as np
 for k, exp_data in experiments.items():
     times = np.array(exp_data['times'])
     avg = np.average(times)
     mse = np.sum(np.square(times - avg))
     csv = exp_data['data']   [avg, mse, avg/exp_data['data'][3], avg/exp_data['data'][3]/exp_data['data'][7]]   exp_data['times']
     print(",".join([str(x) for x in csv]))