diff --git a/code_jurica/loader.py b/code_jurica/loader.py
index 406144fa90e74f9eb3b1f09c550ef8f2ae027daa..ad3009c17d55127258a66e50c663026978c5fb44 100644
--- a/code_jurica/loader.py
+++ b/code_jurica/loader.py
@@ -144,4 +144,4 @@ with open('models/train_test_split_extended.p', 'wb') as handle:
 #     len(labels_train),
 #     len(labels_val)
 #     )
-# )
\ No newline at end of file
+# )
diff --git a/code_jurica/loader_tmp.py b/code_jurica/loader_tmp.py
index 1e0190b285831b3b6c4c299943009456640e9945..84ad20142858fdc6deb4da7ccd998cd8f7122951 100644
--- a/code_jurica/loader_tmp.py
+++ b/code_jurica/loader_tmp.py
@@ -9,13 +9,7 @@ from keras.preprocessing.text import Tokenizer
 from keras.layers import Embedding
 
 #REPRODUCIBLE
-<<<<<<< HEAD
 os.environ['PYTHONHASHSEED'] = '0'
-=======
-from util import TokenizePreprocessor
-
-
->>>>>>> origin/master
 np.random.seed(42)
 random.seed(12345)
 
@@ -27,11 +21,9 @@ SEED = 777
 
 itCorpora, itErrors = prepareData.prepareData('IT')
 huCorpora, huErrors = prepareData.prepareData('HU')
-<<<<<<< HEAD
-=======
 frCorpora, frErrors = prepareData.prepareData('FR')
 # print(len(frErrors), len(itErrors), len(huErrors))
->>>>>>> origin/master
+
 
 try:
     df = pd.DataFrame(frErrors+itErrors+huErrors, columns=['Dataset','DocID', 'MissingRowID'])
diff --git a/code_jurica/seq2seq.py b/code_jurica/seq2seq.py
index 4fd324a4280f14e3e50aeff1bbf248864a385489..9aef90ccd207ad30917cacb548d6ef165207b014 100644
--- a/code_jurica/seq2seq.py
+++ b/code_jurica/seq2seq.py
@@ -12,6 +12,7 @@ import tqdm
 
 import pickle
 from sklearn.metrics import classification_report
+from seq2seq_base import run_pipeline_prediction
 
 #REPRODUCIBLE
 np.random.seed(42)
@@ -183,42 +184,54 @@ def decode_seq(inp_seq):
 
     return translated_sent[:-1], translated_index[:-1]
 
-y_true = []
-y_pred = []
-
-
-source_val = source_kerasTokenizer.texts_to_sequences(source_val)
-source_val = pad_sequences(source_val, maxlen=source_max_sequence_tokenizer, padding='post')
-
-for seq_index in tqdm.tqdm(range(len(source_val))):
-# for seq_index in range(10):
-    inp_seq = source_val[seq_index:seq_index+1]
-    translated_sent, translated_index= decode_seq(inp_seq)
-
-    # PREDICT ICD10
-    source_word_sequence = icd10Tokenizer.texts_to_sequences([" ".join(translated_sent)])
-    word_sequence = pad_sequences(source_word_sequence, maxlen=icd10_model.layers[0].input_shape[1], padding='post')
-    icd10_code_index = icd10_model.predict(word_sequence)
-    # print(icd10_code_index, type(icd10_code_index))
-    max_val_index = np.argmax(icd10_code_index, axis=1)[0]
-    # print(max_val_index)
-    icd10_label = encoded_Y.inverse_transform(max_val_index)
-
-    # print('-')
-    # target_index = np.trim_zeros(target_val[seq_index], 'b')[1:-1]
-    # print('Target indexes:', target_index)
-    # print('Decoded indexes:', translated_index)
-    #
-    # print('Target sentence:', " ".join([target_index_to_word_dict[x] for x in target_index]))
-    # print('Decoded sentence:', " ".join(translated_sent))
-    #
-    # print('Target ICD-10:', labels_val[seq_index])
-    # print('Predict ICD-10:', icd10_label)
-
-    y_true.append(labels_val[seq_index])
-    y_pred.append(icd10_label)
-
-report = classification_report(y_true, y_pred)
-report_df = report_to_df(report)
-report_df.to_csv('logs/classification_report_extended.csv')
-print(report_df)
\ No newline at end of file
+
+max_icd10_length = icd10_model.layers[0].input_shape[1]
+
+run_pipeline_prediction(source_val, decode_seq, icd10_model, encoded_Y, labels_val,
+                        source_kerasTokenizer, source_max_sequence_tokenizer,
+                        icd10Tokenizer, max_icd10_length,
+                        target_index_to_word_dict, target_val,
+                        'logs/seq2seq')
+
+
+
+
+# y_true = []
+# y_pred = []
+#
+#
+# source_val = source_kerasTokenizer.texts_to_sequences(source_val)
+# source_val = pad_sequences(source_val, maxlen=source_max_sequence_tokenizer, padding='post')
+#
+# for seq_index in tqdm.tqdm(range(len(source_val))):
+# # for seq_index in range(10):
+#     inp_seq = source_val[seq_index:seq_index+1]
+#     translated_sent, translated_index= decode_seq(inp_seq)
+#
+#     # PREDICT ICD10
+#     source_word_sequence = icd10Tokenizer.texts_to_sequences([" ".join(translated_sent)])
+#     word_sequence = pad_sequences(source_word_sequence, maxlen=icd10_model.layers[0].input_shape[1], padding='post')
+#     icd10_code_index = icd10_model.predict(word_sequence)
+#     # print(icd10_code_index, type(icd10_code_index))
+#     max_val_index = np.argmax(icd10_code_index, axis=1)[0]
+#     # print(max_val_index)
+#     icd10_label = encoded_Y.inverse_transform(max_val_index)
+#
+#     # print('-')
+#     # target_index = np.trim_zeros(target_val[seq_index], 'b')[1:-1]
+#     # print('Target indexes:', target_index)
+#     # print('Decoded indexes:', translated_index)
+#     #
+#     # print('Target sentence:', " ".join([target_index_to_word_dict[x] for x in target_index]))
+#     # print('Decoded sentence:', " ".join(translated_sent))
+#     #
+#     # print('Target ICD-10:', labels_val[seq_index])
+#     # print('Predict ICD-10:', icd10_label)
+#
+#     y_true.append(labels_val[seq_index])
+#     y_pred.append(icd10_label)
+#
+# report = classification_report(y_true, y_pred)
+# report_df = report_to_df(report)
+# report_df.to_csv('logs/classification_report_extended.csv')
+# print(report_df)
diff --git a/code_jurica/seq2seq_attention.py b/code_jurica/seq2seq_attention.py
index 607409859290739da8db3fc589352382c9b472a6..23bfd073f7967d2b1ad3cda90dcf0c7f5fcaa75b 100644
--- a/code_jurica/seq2seq_attention.py
+++ b/code_jurica/seq2seq_attention.py
@@ -12,6 +12,7 @@ import tqdm
 
 import pickle
 from sklearn.metrics import classification_report
+from seq2seq_base import run_pipeline_prediction
 
 #REPRODUCIBLE
 np.random.seed(42)
@@ -230,42 +231,50 @@ def decode_seq(inp_seq):
 
     return translated_sent[:-1], translated_index[:-1]
 
-y_true = []
-y_pred = []
-
-
-source_val = source_kerasTokenizer.texts_to_sequences(source_val)
-source_val = pad_sequences(source_val, maxlen=source_max_sequence_tokenizer, padding='post')
-
-for seq_index in tqdm.tqdm(range(len(source_val))):
-# for seq_index in range(10):
-    inp_seq = source_val[seq_index:seq_index+1]
-    translated_sent, translated_index= decode_seq(inp_seq)
-
-    # PREDICT ICD10
-    source_word_sequence = icd10Tokenizer.texts_to_sequences([" ".join(translated_sent)])
-    word_sequence = pad_sequences(source_word_sequence, maxlen=icd10_model.layers[0].input_shape[1], padding='post')
-    icd10_code_index = icd10_model.predict(word_sequence)
-    # print(icd10_code_index, type(icd10_code_index))
-    max_val_index = np.argmax(icd10_code_index, axis=1)[0]
-    # print(max_val_index)
-    icd10_label = encoded_Y.inverse_transform(max_val_index)
-
-    # print('-')
-    # target_index = np.trim_zeros(target_val[seq_index], 'b')[1:-1]
-    # print('Target indexes:', target_index)
-    # print('Decoded indexes:', translated_index)
-    #
-    # print('Target sentence:', " ".join([target_index_to_word_dict[x] for x in target_index]))
-    # print('Decoded sentence:', " ".join(translated_sent))
-    #
-    # print('Target ICD-10:', labels_val[seq_index])
-    # print('Predict ICD-10:', icd10_label)
-
-    y_true.append(labels_val[seq_index])
-    y_pred.append(icd10_label)
-
-report = classification_report(y_true, y_pred)
-report_df = report_to_df(report)
-report_df.to_csv('logs/classification_report_extended.csv')
-print(report_df)
+max_icd10_length = icd10_model.layers[0].input_shape[1]
+
+run_pipeline_prediction(source_val, decode_seq, icd10_model, encoded_Y, labels_val,
+                        source_kerasTokenizer, source_max_sequence_tokenizer,
+                        icd10Tokenizer, max_icd10_length,
+                        target_index_to_word_dict, target_val,
+                        'logs/seq2seq')
+
+# y_true = []
+# y_pred = []
+#
+#
+# source_val = source_kerasTokenizer.texts_to_sequences(source_val)
+# source_val = pad_sequences(source_val, maxlen=source_max_sequence_tokenizer, padding='post')
+#
+# for seq_index in tqdm.tqdm(range(len(source_val))):
+# # for seq_index in range(10):
+#     inp_seq = source_val[seq_index:seq_index+1]
+#     translated_sent, translated_index= decode_seq(inp_seq)
+#
+#     # PREDICT ICD10
+#     source_word_sequence = icd10Tokenizer.texts_to_sequences([" ".join(translated_sent)])
+#     word_sequence = pad_sequences(source_word_sequence, maxlen=icd10_model.layers[0].input_shape[1], padding='post')
+#     icd10_code_index = icd10_model.predict(word_sequence)
+#     # print(icd10_code_index, type(icd10_code_index))
+#     max_val_index = np.argmax(icd10_code_index, axis=1)[0]
+#     # print(max_val_index)
+#     icd10_label = encoded_Y.inverse_transform(max_val_index)
+#
+#     # print('-')
+#     # target_index = np.trim_zeros(target_val[seq_index], 'b')[1:-1]
+#     # print('Target indexes:', target_index)
+#     # print('Decoded indexes:', translated_index)
+#     #
+#     # print('Target sentence:', " ".join([target_index_to_word_dict[x] for x in target_index]))
+#     # print('Decoded sentence:', " ".join(translated_sent))
+#     #
+#     # print('Target ICD-10:', labels_val[seq_index])
+#     # print('Predict ICD-10:', icd10_label)
+#
+#     y_true.append(labels_val[seq_index])
+#     y_pred.append(icd10_label)
+#
+# report = classification_report(y_true, y_pred)
+# report_df = report_to_df(report)
+# report_df.to_csv('logs/classification_report_extended.csv')
+# print(report_df)
diff --git a/code_jurica/seq2seq_base.py b/code_jurica/seq2seq_base.py
new file mode 100644
index 0000000000000000000000000000000000000000..ca5ff4b9404659b3604f731a01951f9d36d14c0f
--- /dev/null
+++ b/code_jurica/seq2seq_base.py
@@ -0,0 +1,100 @@
+import tqdm
+import numpy as np
+import os
+from typing import List, Callable, Dict, Iterable
+
+from keras import Model
+from keras.preprocessing.sequence import pad_sequences
+from keras.preprocessing.text import Tokenizer
+from sklearn.metrics import classification_report
+from sklearn.preprocessing import LabelEncoder
+
+from util import report_to_df
+
+
+def run_pipeline_prediction(sentences: List[str], decode_seq_fnc: Callable, icd10_model: Model,
+                            icd10_encoder: LabelEncoder, gold_labels: List,
+                            source_tokenizer: Tokenizer, max_source_length: int,
+                            icd10_tokenizer: Tokenizer, max_icd10_length: int,
+                            target_index_to_word_dict: Dict, gold_target_indexes,
+                            log_dir: str):
+
+    os.makedirs(log_dir, exist_ok=True)
+
+    y_true = []
+    y_pred = []
+
+    source_val = source_tokenizer.texts_to_sequences(sentences)
+    source_val = pad_sequences(source_val, maxlen=max_source_length, padding='post')
+
+    pred_texts = []
+    pred_ids = []
+
+    gold_texts = []
+    gold_ids = []
+
+    for seq_index in tqdm.tqdm(range(len(source_val)), total=len(source_val)):
+        # for seq_index in range(10):
+        inp_seq = source_val[seq_index:seq_index + 1]
+        translated_sent, translated_index = decode_seq_fnc(inp_seq)
+
+        # PREDICT ICD10
+        source_word_sequence = icd10_tokenizer.texts_to_sequences([" ".join(translated_sent)])
+        word_sequence = pad_sequences(source_word_sequence, maxlen=max_icd10_length, padding='post')
+        icd10_code_index = icd10_model.predict(word_sequence)
+        # print(icd10_code_index, type(icd10_code_index))
+        max_val_index = np.argmax(icd10_code_index, axis=1)[0]
+
+        # print(max_val_index)
+        icd10_label = icd10_encoder.inverse_transform(max_val_index)
+
+        y_true.append(gold_labels[seq_index])
+        y_pred.append(icd10_label)
+
+        # Debugging prediction
+        pred_text = " ".join(translated_sent)
+        pred_texts.append(pred_text)
+
+        pred_indexes = " ".join(translated_index)
+        pred_ids.append(pred_indexes)
+
+        gold_indexes = np.trim_zeros(gold_target_indexes[seq_index], 'b')[1:-1]
+        gold_ids.append(" ".join(gold_indexes))
+
+        gold_text = " ".join([target_index_to_word_dict[x] for x in gold_indexes])
+        gold_texts.append(gold_text)
+
+        print('Target indexes:', gold_indexes)
+        print('Decoded indexes:', pred_indexes)
+
+        print('Target text:', gold_text)
+        print('Decoded text:', pred_text)
+
+        print('Target ICD-10:', gold_labels[seq_index])
+        print('Predict ICD-10:', icd10_label)
+
+        print('----------------------------------------------')
+
+
+    # Write classication report
+    report = classification_report(y_true, y_pred)
+    report_df = report_to_df(report)
+    report_file = os.path.join(log_dir, 'classification_report.csv')
+    report_df.to_csv(report_file)
+    print(report_df)
+
+    # Save prediction debugging information
+    save_list_to_file(pred_texts, os.path.join(log_dir, 'pred_texts.txt'))
+    save_list_to_file(pred_ids, os.path.join(log_dir, 'pred_ids.txt'))
+
+    save_list_to_file(gold_texts, os.path.join(log_dir, 'gold_texts.txt'))
+    save_list_to_file(gold_ids, os.path.join(log_dir, 'gold_ids.txt'))
+
+
+def save_list_to_file(values: Iterable, output_file: str):
+    with open(output_file, "w", encoding="utf-8") as output_writer:
+        output_content = ["%s\n" % value for value in values]
+        output_writer.writelines(output_content)
+        output_writer.close()
+
+
diff --git a/code_jurica/test.py b/code_jurica/test.py
index 954c8fdcfa2245af9877de41bba4beba0db0aae2..11808bc4c8d811e73a32bab492621e9d1cfb3046 100644
--- a/code_jurica/test.py
+++ b/code_jurica/test.py
@@ -169,4 +169,4 @@ for k,v in runs.items():
 
     report = classification_report(y_true, y_pred)
     report_df = report_to_df(report)
-    report_df.to_csv('logs/classification_report_test{}.csv'.format(v['s2s']))
\ No newline at end of file
+    report_df.to_csv('logs/classification_report_test{}.csv'.format(v['s2s']))
diff --git a/code_mario/app_context.py b/code_mario/app_context.py
index 772341228320c615fba32b2bc1c9d654ffd5df29..a32ca22ca7d8c742ee35d3befa7e46382564f836 100644
--- a/code_mario/app_context.py
+++ b/code_mario/app_context.py
@@ -1,5 +1,5 @@
 import os
-from util import LogUtil
+from common_util import LogUtil
 
 
 class AppContext(object):
diff --git a/code_mario/clef18_task1_base.py b/code_mario/clef18_task1_base.py
index d94136309f92acc71118e34a11e117b7b9fa8d19..8303bdd6a07368015ecfb0517bd71b97ab116301 100644
--- a/code_mario/clef18_task1_base.py
+++ b/code_mario/clef18_task1_base.py
@@ -21,7 +21,7 @@ from sklearn.externals import joblib
 from sklearn.model_selection import train_test_split
 
 from app_context import AppContext
-from util import LoggingMixin
+from common_util import LoggingMixin
 from dnn_classifiers import NeuralNetworkClassifiers as nnc
 from keras_extension import KerasUtil as ku
 
diff --git a/code_mario/clef18_task1_data.py b/code_mario/clef18_task1_data.py
index d24c43a05c101cf8d35a11651d5726dfd9b8ad49..8bacd54ff036147f05eb31ea283a96280c688a11 100644
--- a/code_mario/clef18_task1_data.py
+++ b/code_mario/clef18_task1_data.py
@@ -7,7 +7,7 @@ from pandas import DataFrame
 from tqdm import tqdm
 
 from app_context import AppContext
-from util import LoggingMixin
+from common_util import LoggingMixin
 from itertools import groupby
 from preprocessing import DataPreparationUtil as pdu
 
diff --git a/code_mario/util.py b/code_mario/common_util.py
similarity index 100%
rename from code_mario/util.py
rename to code_mario/common_util.py
diff --git a/code_mario/ft_embeddings.py b/code_mario/ft_embeddings.py
index 47fb9b5696038298f7cab8331d1172e8eb034264..57c18c0390bb7401064e1a664c3334e5104d6def 100644
--- a/code_mario/ft_embeddings.py
+++ b/code_mario/ft_embeddings.py
@@ -4,7 +4,7 @@ from typing import List, Dict
 from gensim.models import FastText
 
 from app_context import AppContext
-from util import LoggingMixin
+from common_util import LoggingMixin
 
 
 class FastTextModel(LoggingMixin):
diff --git a/code_mario/keras_extension.py b/code_mario/keras_extension.py
index 81cbb644158f1b441074a9e920d34873b2b2b58b..0ee4c9e3c18518b2f358c32f91a22f58f9359630 100644
--- a/code_mario/keras_extension.py
+++ b/code_mario/keras_extension.py
@@ -6,7 +6,7 @@ from keras.callbacks import Callback, ModelCheckpoint, CSVLogger, EarlyStopping
 from keras.wrappers.scikit_learn import KerasClassifier
 
 from app_context import AppContext
-from util import LoggingMixin
+from common_util import LoggingMixin
 
 
 class KerasUtil(object):
diff --git a/code_mario/preprocessing.py b/code_mario/preprocessing.py
index 45b540db2fbe180913d1eefaea1dadbcb2290951..963aed4395b1079e2343f050631e895a15099d15 100644
--- a/code_mario/preprocessing.py
+++ b/code_mario/preprocessing.py
@@ -8,7 +8,7 @@ from sklearn.base import BaseEstimator, TransformerMixin
 from sklearn.preprocessing import LabelEncoder
 from typing import Callable, List
 
-from util import PandasUtil
+from common_util import PandasUtil
 
 
 class DataPreparationUtil(object):