Make it simpler

35-dnn-no-learning/normalize-chars.py

@@ -2,7 +2,6 @@ from keras.models import Model
 from keras import layers
 from keras.layers import Input, Dense
 from keras.utils import plot_model
-
 import numpy as np
 import sys, os, string
 
@@ -14,10 +13,10 @@ INPUT_VOCAB_SIZE = len(characters)
 LINE_SIZE = 100
 
 def encode_one_hot(s):
-    """One-hot encode all characters of the given string.
-    """
     all = []
     for c in s:
+        if c not in characters:
+            continue
         x = np.zeros((INPUT_VOCAB_SIZE)) 
         index = char_indices[c]
         x[index] = 1 
@@ -25,13 +24,11 @@ def encode_one_hot(s):
     return all
 
 def decode_one_hot(x):
-    """Return a string from a one-hot-encoded matrix
-    """
     s = []
     for onehot in x:
-        one_index = np.where(onehot == 1) # one_index is a tuple of two things
-        if len(one_index[0]) > 0:
-            n = one_index[0][0]
+        one_index = np.where(onehot == 1) # tuple of two things
+        if len(one_index[1]) > 0:
+            n = one_index[1][0]
             c = indices_char[n]
             s.append(c) 
     return ''.join(s)
@@ -60,11 +57,8 @@ def normalization_layer_set_weights(n_layer):
     n_layer.set_weights(wb)
     return n_layer
 
-
 def build_model():
-    print('Build model...')
-    
-    # Normalize every character in the input, using a shared dense model
+    # Normalize characters using a shared dense model
     n_layer = Dense(INPUT_VOCAB_SIZE)
     raw_inputs = []
     normalized_outputs = []
@@ -75,17 +69,10 @@ def build_model():
         normalized_outputs.append(filtered_char)
     normalization_layer_set_weights(n_layer)
 
-    merged_output = layers.concatenate(normalized_outputs, axis=-1)
-
-    reshape = layers.Reshape((LINE_SIZE, INPUT_VOCAB_SIZE, ))
-    reshaped_output = reshape(merged_output)
-
-    model = Model(inputs=raw_inputs, outputs=reshaped_output)
-
+    model = Model(inputs=raw_inputs, outputs=normalized_outputs)
     return model
 
 model = build_model()
-#model.summary()
 plot_model(model, to_file='normalization.png', show_shapes=True)
 
 with open(sys.argv[1]) as f:
@@ -100,9 +87,7 @@ with open(sys.argv[1]) as f:
             data[j].append(np.zeros((INPUT_VOCAB_SIZE)))
 
         inputs = [np.array(e) for e in data]
-
         preds = model.predict(inputs)
-        normal = decode_one_hot(preds[0])
+        normal = decode_one_hot(preds)
 
-#        print(decode_one_hot(onehots))
         print(normal)

35-dnn-no-learning/tf-35.py

@@ -34,8 +34,8 @@ def decode_one_hot(x):
     s = []
     for onehot in x:
         one_index = np.where(onehot == 1) # one_index is a tuple of two things
-        if len(one_index[1]) > 0:
-            n = one_index[1][0]
+        if len(one_index[0]) > 0:
+            n = one_index[0][0]
             c = indices_char[n]
             s.append(c) 
     return ''.join(s)
@@ -119,7 +119,7 @@ def build_model():
     # Find the space characters
     words_output = layers.Lambda(SpaceDetector)(reshaped_output)
 
-    model = Model(inputs=raw_inputs, outputs=normalized_outputs)
+    model = Model(inputs=raw_inputs, outputs=words_output)
 
     return model