Manually merged in commits that cleaned up the epoch count and how to

loop through the active frames and use the sampling density parameter
from the frameworks branch into the release_01 branch.

Author: bvanessen

Pilot2/P2B1/p2b1.py

@@ -40,9 +40,9 @@
 {'name':'set_sel','help':'[3k_Disordered, 3k_Ordered, 3k_Ordered_and_gel, 6k_Disordered, 6k_Ordered, 6k_Ordered_and_gel]','type':str,'default':'3k_Disordered'},
 {'name':'conv_bool', 'type':candle.str2bool, 'default':True, 'help':'Invoke training using 1D Convs for inner AE'},
 {'name':'full_conv_bool', 'type':candle.str2bool, 'default':False, 'help':'Invoke training using fully convolutional NN for inner AE'},
-{'name':'type_bool', 'type':candle.str2bool, 'default':False, 'help':'Include molecule type information in desining AE'},
+{'name':'type_bool', 'type':candle.str2bool, 'default':True, 'help':'Include molecule type information in desining AE'},
 {'name':'nbr_type', 'type':str, 'default':'relative', 'help':'Defines the type of neighborhood data to use. [relative, invariant]'},
-{'name':'backend', 'help':'Keras Backend', 'type':str, 'default':'theano'}
+{'name':'backend', 'help':'Keras Backend', 'type':str, 'default':'tensorflow'}
 ]
 
 required = [
@@ -58,7 +58,6 @@
     # note 'cool' is a boolean
     'cool',
 
-    'molecular_epochs',
     'molecular_num_hidden',
     'molecular_nonlinearity',
     'molecular_nbrs',
@@ -267,7 +266,11 @@ def datagen(self, epoch=0, print_out=1, test=0):
             xt_all = np.array([])
             yt_all = np.array([])
 
-            for i in range(num_frames):
+            num_active_frames = random.sample(range(num_frames), int(self.sampling_density*num_frames))
+
+            print('Datagen on the following frames', num_active_frames)
+
+            for i in num_active_frames:
 
                 if self.conv_net:
                     xt = Xnorm[i]
@@ -308,7 +311,7 @@ def datagen(self, epoch=0, print_out=1, test=0):
 
     def train_ac(self):
 
-        for i in range(self.mb_epochs):
+        for i in range(1, self.mb_epochs+1):
             print ("\nTraining epoch: {:d}\n".format(i))
 
             frame_loss = []
@@ -322,8 +325,8 @@ def train_ac(self):
             encoder_weight_file = '%s/%s.hdf5' % (current_path, 'encoder_weights')
 
             for curr_file, xt_all, yt_all in self.datagen(i):
-                for frame in random.sample(range(len(xt_all)), int(self.sampling_density*len(xt_all))):
-
+                #for frame in random.sample(range(len(xt_all)), int(self.sampling_density*len(xt_all))):
+                for frame in range(len(xt_all)):
                     history = self.molecular_model.fit(xt_all[frame], yt_all[frame], epochs=1,
                                                        batch_size=self.batch_size, callbacks=self.callbacks[:2],
                                                        verbose=0)

Pilot2/P2B1/p2b1_baseline_keras2.py

@@ -28,7 +28,6 @@
 
 HOME = os.environ['HOME']
 
-
 def parse_list(option, opt, value, parser):
     setattr(parser.values, option.dest, value.split(','))
 
@@ -163,7 +162,7 @@ def run(GP):
         type_feat_vect = []
     else:
         num_type_features = 5
-        type_feat_vect = fields.keys()[3:8]
+        type_feat_vect = list(fields.keys())[3:8]
 
     num_features = num_loc_features + num_type_features + num_beads
     dim = np.prod([num_beads, num_features, molecular_nbrs+1])
@@ -226,7 +225,7 @@ def run(GP):
     molecular_model.summary()
     ##### set up callbacks and cooling for the molecular_model ##########
     drop = 0.5
-    mb_epochs = GP['molecular_epochs']
+    mb_epochs = GP['epochs']
     initial_lrate = GP['learning_rate']
     epochs_drop = 1+int(np.floor(mb_epochs/3))
 

Pilot2/P2B1/p2b1_default_model.txt

@@ -12,9 +12,8 @@ activation='relu'
 molecular_nonlinearity='elu'
 molecular_num_hidden=[256, 128, 64, 32, 16, 8]
 molecular_nbrs = 10
-molecular_epochs=1
 base_memo='p2b1'
 drop_prob = 0.5
 data_set='3k_Ordered'
-sampling_density = 0.01
+sampling_density = 0.05
 save_path='.'

README.setup.linux

@@ -22,6 +22,7 @@ conda install -c conda-forge keras=2
 conda install -c anaconda pandas
 conda install -c anaconda scikit-learn 
 conda install -c anaconda matplotlib
+---conda install -c conda-forge pygpu
 
 # Install additional modules for Pilot2 benchmarks
 conda install -c conda-forge opencv