0.65, it doesnt get better

fromNotebook.py

@@ -16,7 +16,7 @@
 
 
 trainingdata = pd.read_csv(path_data + fn_train, usecols=['CITY', 'INSTITUTION_FULL','ADDRESS_FULL','d_state']) #,'OBSERVATION'
-trainingdata = trainingdata[0:1500] 
+trainingdata = trainingdata[0:2200] 
 #trainingdata.head(10)
 
 
@@ -26,7 +26,8 @@
 
 print(df.head(9))
 df.to_csv('merged.csv')
-df.reindex(np.random.permutation(df.index)) # shuffle the rows, because I suspect some row-dependence...
+
+df.reindex(np.random.permutation(df.index)) # shuffle the rows, because I suspect some row-dependence... undo for debugging
 
 # prepare data for keras:
 
@@ -42,15 +43,15 @@
 #test_files_names = dtrainingata['filename'][train_size:]
 
 # 32 states and NA
-num_labels = 23   # there are only 25 states with institiutions in the dataset
-vocab_size = 3000  # how many different words to keep?
+num_labels = 26   # there are only 25 states with institiutions in the dataset
+vocab_size = 23000  # how many different words to keep?
 batch_size = 150
 
 from keras.preprocessing.text import Tokenizer
 from keras.models import Sequential
-from keras.layers import Activation, Dense, Dropout
+from keras.layers import Activation, Dense, Dropout, Conv1D, MaxPooling1D,GlobalMaxPool1D
 from keras.constraints import maxnorm
-from keras.utils import plot_model
+from keras.utils import plot_model, to_categorical
 
 
 #print(train_input.head(10))
@@ -77,14 +78,27 @@
 y_test = encoder.transform(test_tags)
 
 
+#print("train tags_ ", train_tags[0:20])
+#y_train = to_categorical(train_tags, num_labels)
+#y_test = to_categorical(test_tags, num_labels)
+
+text_labels = encoder.classes_
+
+print("y_train: ", y_train[0:10])
+for ls in range(10):
+    print("y_train argmax",(np.argmax(y_train[ls])))
+print("y_test",  y_test[0:10])
+print("labels: ", text_labels)
 #exit()
 
 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(Activation('relu'))
-model.add(Dropout(0.4))
+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))