Add channel ordering option + fix for k2 bug

MPIGDriver.py

@@ -156,7 +156,7 @@
                 mode='easgd', sync_every=args.sync_every,
                 worker_optimizer=args.worker_optimizer,
                 worker_optimizer_params=args.worker_optimizer_params,
-                elastic_force=args.elastic_force/(comm.Get_size()-1),
+                elastic_force=args.elastic_force/(min(1,comm.Get_size()-1)),
                 elastic_lr=args.elastic_lr, 
                 elastic_momentum=args.elastic_momentum) 
     else:

models/get_3d.py

@@ -7,7 +7,7 @@
     print ("hum")
 import numpy as np
 import sys
-
+import keras
 def get_data(datafile):
     #get data for training
     #print ('Loading Data from .....', datafile)
@@ -20,7 +20,11 @@ def get_data(datafile):
     X = X.astype(np.float32)
     y = y.astype(np.float32)
     y = y/100.
-    ecal = np.squeeze(np.sum(X, axis=(1, 2, 3)))
+    if keras.backend.image_data_format() !='channels_last':
+       X =np.moveaxis(X, -1, 1)
+       ecal = np.squeeze(np.sum(X, axis=(2, 3, 4)))
+    else:
+       ecal = np.squeeze(np.sum(X, axis=(1, 2, 3)))
     print (X.shape)
     print (y.shape)
     print (ecal.shape)

mpi_learn/train/GanModel.py

@@ -114,8 +114,16 @@ def _Model(**args):
     else:
         return Model(**args)
 def discriminator(fixed_bn = False, discr_drop_out=0.2):
+    if keras.backend.image_data_format() =='channels_last':
+        dshape=(25, 25, 25,1)
+        daxis=(1,2,3)
+    else:
+        dshape=(1, 25, 25, 25)
+        daxis=(2,3,4)
+
+    image = Input(shape=dshape, name='image')
+
 
-    image = Input(shape=( 25, 25, 25,1 ), name='image')
 
     bnm=2 if fixed_bn else 0
     f=(5,5,5)
@@ -164,19 +172,21 @@ def discriminator(fixed_bn = False, discr_drop_out=0.2):
 
     fake = _Dense(1, activation='sigmoid', name='classification')(dnn_out)
     aux = _Dense(1, activation='linear', name='energy')(dnn_out)
-    ecal = Lambda(lambda x: K.sum(x, axis=(1, 2, 3)), name='sum_cell')(image)
+    ecal = Lambda(lambda x: K.sum(x, daxis), name='sum_cell')(image)
 
     return _Model(output=[fake, aux, ecal], input=image, name='discriminator_model')
 
 def generator(latent_size=200, return_intermediate=False, with_bn=True):
-
+    if keras.backend.image_data_format() =='channels_last':
+        dim = (7,7,8,8)
+    else:
+        dim = (8, 7, 7,8)
     latent = Input(shape=(latent_size, ))
-
     bnm=0
     x = _Dense(64 * 7* 7, init='glorot_normal',
               name='gen_dense1'
     )(latent)
-    x = Reshape((7, 7,8, 8))(x)
+    x = Reshape(dim)(x)
     x = _Conv3D(64, 6, 6, 8, border_mode='same', init='he_uniform',
                name='gen_c1'
     )(x)
@@ -212,9 +222,14 @@ def generator(latent_size=200, return_intermediate=False, with_bn=True):
     return _Model(input=[latent], output=fake_image, name='generator_model')
 
 def get_sums(images):
-    sumsx = np.squeeze(np.sum(images, axis=(2,3)))
-    sumsy = np.squeeze(np.sum(images, axis=(1,3)))
-    sumsz = np.squeeze(np.sum(images, axis=(1,2)))
+    if keras.backend.image_data_format() =='channels_last':
+       sumsx = np.squeeze(np.sum(images, axis=(2,3)))
+       sumsy = np.squeeze(np.sum(images, axis=(1,3)))
+       sumsz = np.squeeze(np.sum(images, axis=(1,2)))
+    else: 
+       sumsx = np.squeeze(np.sum(images, axis=(3,4)))
+       sumsy = np.squeeze(np.sum(images, axis=(2,4)))
+       sumsz = np.squeeze(np.sum(images, axis=(2,3)))
     return sumsx, sumsy, sumsz
 
 def get_moments(images, sumsx, sumsy, sumsz, totalE, m):
@@ -535,6 +550,9 @@ def make_opt(**args):
             loss=['binary_crossentropy', 'mean_absolute_percentage_error', 'mean_absolute_percentage_error'],
             loss_weights=self.discr_loss_weights
         )
+        if kv2: 
+           self.discriminator.trainable = True #workaround for keras 2 bug
+
         self.combined.metrics_names = self.discriminator.metrics_names
         print (self.discriminator.metrics_names)
         print (self.combined.metrics_names)