implemented more general hash that allows deep copies of the conf to maintain the hash

Julian Kates-Harbeck · Julian Kates-Harbeck · commit 8aa5c1ae7cd0 · 2018-03-20T20:29:51.000-04:00
diff --git a/plasma/models/builder.py b/plasma/models/builder.py
@@ -19,7 +19,7 @@
 import os,sys
 import numpy as np
 from copy import deepcopy
-from plasma.utils.downloading import makedirs_process_safe
+from plasma.utils.downloading import makedirs_process_safe,general_object_hash
 
 import hashlib
 
@@ -31,18 +31,22 @@ def on_batch_end(self, batch, logs=None):
         self.losses.append(logs.get('loss'))
 
 
+
 class ModelBuilder(object):
     def __init__(self,conf):
         self.conf = conf
 
     def get_unique_id(self):
-        num_epochs = self.conf['training']['num_epochs']
+        #num_epochs = self.conf['training']['num_epochs']
         this_conf = deepcopy(self.conf)
-        #don't make hash dependent on number of epochs.
+        #don't make hash dependent on number of epochs or T_min_warn as those can be modified
         this_conf['training']['num_epochs'] = 0
-        unique_id = int(hashlib.md5((dill.dumps(this_conf).decode('unicode_escape')).encode('utf-8')).hexdigest(),16)
+        this_conf['data']['T_min_warn'] = 30
+        #unique_id = int(hashlib.md5((dill.dumps(this_conf).decode('unicode_escape')).encode('utf-8')).hexdigest(),16)
+        unique_id = general_object_hash(this_conf)
         return unique_id
 
+
     def get_0D_1D_indices(self):
         #make sure all 1D indices are contiguous in the end!
         use_signals = self.conf['paths']['use_signals']
@@ -270,6 +274,8 @@ def get_all_saved_files(self):
         self.ensure_save_directory()
         unique_id = self.get_unique_id()
         filenames = os.listdir(self.conf['paths']['model_save_path'])
+        print("All saved files with id {} and path {}".format(unique_id,self.conf['paths']['model_save_path']))
+        print(filenames)
         epochs = []
         for file in filenames:
             curr_id,epoch = self.extract_id_and_epoch_from_filename(file)
diff --git a/plasma/models/mpi_runner.py b/plasma/models/mpi_runner.py
@@ -731,6 +731,7 @@ def mpi_train(conf,shot_list_train,shot_list_validate,loader, callbacks_list=Non
             mpi_model.batch_iterator_func.__exit__()
             mpi_model.num_so_far_accum = mpi_model.num_so_far_indiv
             mpi_model.set_batch_iterator_func()
+
         if 'monitor_test' in conf['callbacks'].keys() and conf['callbacks']['monitor_test']:
             conf_curr = deepcopy(conf)
             T_min_warn_orig = conf['data']['T_min_warn']
diff --git a/plasma/utils/downloading.py b/plasma/utils/downloading.py
@@ -26,6 +26,8 @@
 import os
 import errno
 
+import dill,hashlib
+
 # import gadata
 
 # from plasma.primitives.shots import ShotList
@@ -34,6 +36,26 @@
 
 #print("Importing numpy version"+np.__version__)
 
+def general_object_hash(o):
+    """Makes a hash from a dictionary, list, tuple or set to any level, that contains
+       only other hashable types (including any lists, tuples, sets, and
+       dictionaries). Relies on dill for serialization"""
+         
+    if isinstance(o, (set, tuple, list)):
+        return tuple([make_hash(e) for e in o])    
+         
+    elif not isinstance(o, dict):
+        return myhash(o)
+    
+    new_o = deepcopy(o)
+    for k, v in new_o.items():
+        new_o[k] = make_hash(v)
+ 
+    return myhash(tuple(frozenset(sorted(new_o.items()))))
+
+def myhash(x):
+    return int(hashlib.md5((dill.dumps(x).decode('unicode_escape')).encode('utf-8')).hexdigest(),16)
+
 
 def get_missing_value_array():
     return np.array([-1.0])
