...works, but only 63%..

fromNotebook.py

@@ -12,7 +12,7 @@
 
 # Source file directory
 path_data = "U:/data/"
-fn_train = "train3.csv" #"filewos_bib_random_nastates.csv"
+fn_train = "train4-womexico.csv" #"train3.csv" #"filewos_bib_random_nastates.csv"
 
 
 trainingdata = pd.read_csv(path_data + fn_train, usecols=['CITY', 'INSTITUTION_FULL','ADDRESS_FULL','d_state']) #,'OBSERVATION'
@@ -44,7 +44,7 @@
 
 # 32 states and NA
 num_labels = 26   # there are only 25 states with institiutions in the dataset
-vocab_size = 23000  # how many different words to keep?
+vocab_size = 16000  # how many different words to keep?
 batch_size = 150
 
 from keras.preprocessing.text import Tokenizer
@@ -94,14 +94,12 @@
 print("\n\n ---------- and now the actual keras training: ----------- \n\n")
 
 model = Sequential()
-model.add(Dense(int(512/2), input_shape=(vocab_size,)))
+model.add(Dense(int(100), input_shape=(vocab_size,)))
 model.add(Activation('relu'))
-model.add(Dropout(0.3))
-#model.add(Conv1D(256, 5, activation='relu'))
-#model.add(GlobalMaxPool1D())
-model.add(Dense(int(256/2)))
-model.add(Activation('relu'))
-model.add(Dropout(0.2))
+model.add(Dropout(0.5))
+#model.add(Dense(int(200)))
+#model.add(Activation('relu'))
+#model.add(Dropout(0.3))
 model.add(Dense(num_labels))
 model.add(Activation('softmax'))
 model.summary()
@@ -112,7 +110,7 @@
 
 history = model.fit(x_train, y_train,
                     batch_size=batch_size,
-                    epochs=15,
+                    epochs=10,
                     verbose=1,
                     validation_split=0.2)