Using two-convolutional layer neural network and setting w0=3000A.

daniel-muthukrishna · May 28, 2018 · 52e72c3 · 52e72c3
1 parent 2f26326
commit 52e72c3
Show file tree

Hide file tree

Showing 5 changed files with 9 additions and 10 deletions.
diff --git a/dash/create_and_save_all_data_files.py b/dash/create_and_save_all_data_files.py
@@ -12,7 +12,7 @@
 
 
 if __name__ == '__main__':
-    modelName = 'zeroZ_trainOnAll_morehostfracs'
+    modelName = 'zeroZ_trainOnAll_w0-3000'
     trainWithHost = True
     classifyHost = False
     minZ = 0.
@@ -38,9 +38,9 @@
         f.write("Num of Redshifts: {}\n".format(numOfRedshifts))
         f.write("Fraction of Training Set Used: {}\n".format(numOfRedshifts))
         f.write("Training Amount: 50 x 500000\n")
-        f.write("Changed wavelength range to 3000 to 10000A\n")
+        f.write("Changed wavelength range to 3500 to 10000A\n")
         f.write("Set outer region to 0.5\n")
-        f.write("Added 3rd convolutional layer to neural network\n")
+        f.write("Using 2 convolutional layers in neural network\n")
         dataFilenames.append(modelInfoFilename)
 
     # CREATE PARAMETERS PICKLE FILE

diff --git a/dash/create_arrays.py b/dash/create_arrays.py
@@ -146,10 +146,13 @@ def shuffle_arrays(self, memmapName='', **kwargs):
                 arrayShuf = np.memmap('shuffled_{}_{}_{}.dat'.format(key, memmapName, self.randnum), dtype=object, mode='w+', shape=arraySize)
             kwargShuf[key] = arrayShuf
 
+        print("Shuffling...")
         # Randomise order
         p = np.random.permutation(len(kwargs['labels']))
         for key in kwargs:
             assert len(kwargs[key]) == arraySize
+            print(key, "shuffling...")
+            print(len(p))
             kwargShuf[key] = kwargs[key][p]
 
         return kwargShuf

diff --git a/dash/deep_learning_multilayer.py b/dash/deep_learning_multilayer.py
@@ -76,11 +76,7 @@ def train_model(dataDirName, overwrite=False):
     a = []
     x, y_, keep_prob, y_conv = convnet_variables(imWidth, imWidthReduc, N, nLabels)
 
-    config = tf.ConfigProto()
-    config.intra_op_parallelism_threads = 44
-    config.inter_op_parallelism_threads = 44
-
-    with tf.Session(config=config) as sess: # config=tf.ConfigProto(inter_op_parallelism_threads=1, intra_op_parallelism_threads=1)) as sess:
+    with tf.Session() as sess: # config=tf.ConfigProto(inter_op_parallelism_threads=1, intra_op_parallelism_threads=1)) as sess:
         # TRAIN AND EVALUATE MODEL
         cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y_conv + 1e-8), reduction_indices=[1]))
         train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

diff --git a/dash/multilayer_convnet.py b/dash/multilayer_convnet.py
@@ -70,7 +70,7 @@ def convnet_variables(imWidth, imWidthReduc, N, ntypes):
     h_fc2 = Layer3.connect_layers(h_pool3, 64, 2)
 
     # READOUT LAYER
-    keep_prob, h_fc2_drop = Layer3.dropout(h_fc2)
+    keep_prob, h_fc2_drop = Layer3.dropout(h_fc1)
     W_fc3, b_fc3 = Layer3.readout_layer()
     y_conv = tf.nn.softmax(tf.matmul(h_fc2_drop, W_fc3) + b_fc3)
 

diff --git a/dash/training_parameters.py b/dash/training_parameters.py
@@ -8,7 +8,7 @@ def create_training_params_file(dataDirName):
                      'Ib-norm', 'Ibn', 'IIb', 'Ib-pec', 'Ic-norm', 'Ic-broad',
                      'Ic-pec', 'IIP', 'IIL', 'IIn', 'II-pec'],
         'nTypes': 17,
-        'w0': 3000.,  # wavelength range in Angstroms
+        'w0': 3500.,  # wavelength range in Angstroms
         'w1': 10000.,
         'nw': 1024,  # number of wavelength bins
         'minAge': -20.,