diff --git a/.gitignore b/.gitignore
index 92a65a9..c26a768 100644
--- a/.gitignore
+++ b/.gitignore
@@ -130,3 +130,6 @@ dmypy.json
 
 # IDE
 .idea/
+
+# Outputs
+output/
diff --git a/KoozDawa/data/apocalypse b/KoozDawa/data/apocalypse.txt
similarity index 100%
rename from KoozDawa/data/apocalypse
rename to KoozDawa/data/apocalypse.txt
diff --git a/KoozDawa/dawa/loader.py b/KoozDawa/dawa/loader.py
index 9ccf9df..682d16c 100644
--- a/KoozDawa/dawa/loader.py
+++ b/KoozDawa/dawa/loader.py
@@ -1,5 +1,7 @@
 import os
 import string
+from pprint import pprint
+from random import choice, randint
 
 from numpy.random import seed
 from tensorflow_core.python.framework.random_seed import set_random_seed
@@ -11,7 +13,9 @@ def load_kawa(root="./"):
     seed(1)
     data_dir = root + 'data/'
     all_lines = []
-    for filename in os.listdir(data_dir):
+    files = os.listdir(data_dir)
+    print("%i files in data folder." % len(files))
+    for filename in files:
         with open(data_dir + filename) as f:
             content = f.readlines()
             all_lines.extend(content)
@@ -23,6 +27,19 @@ def load_kawa(root="./"):
     return all_lines
 
 
+def load_seeds(kawa=None, nb_seeds=10):
+    if kawa is None:
+        kawa = load_kawa()
+    seeds = []
+    for i in range(nb_seeds):
+        plain_kawa = filter(lambda k: k != "\n", kawa)
+        chosen = choice(list(plain_kawa))
+        split = chosen.split(" ")
+        nb_words = randint(1, len(split))
+        seeds.append(split[:nb_words])
+    return seeds
+
+
 def clean_text(lines):
     """
     In dataset preparation step, we will first perform text cleaning of the data
@@ -37,6 +54,8 @@ def main():
     lines = load_kawa("../")
     clean = clean_text(lines)
     print(clean)
+    print("Some seeds:\n\n")
+    pprint(load_seeds(lines))
 
 
 if __name__ == '__main__':
diff --git a/KoozDawa/dawa/lstm.py b/KoozDawa/dawa/lstm.py
index 478fe19..ed2c0f7 100644
--- a/KoozDawa/dawa/lstm.py
+++ b/KoozDawa/dawa/lstm.py
@@ -7,7 +7,7 @@ from keras.utils import to_categorical
 from keras_preprocessing.sequence import pad_sequences
 from keras_preprocessing.text import Tokenizer
 
-from KoozDawa.dawa.loader import load_kawa, clean_text
+from KoozDawa.dawa.loader import load_kawa, clean_text, load_seeds
 from KoozDawa.dawa.tokens import get_sequence_of_tokens
 
 warnings.filterwarnings("ignore")
@@ -61,35 +61,31 @@ def generate_text(model, tokenizer, seed_text="", nb_words=5, max_sequence_len=0
 
 def main():
     should_train = True
-    nb_epoch = 100
-    max_sequence_len = 61  # TODO: Test different default
     # model_file = "../models/dawa_lstm_%i.hd5" % nb_epoch
+    nb_epoch = 100
+    nb_words = 200
     tokenizer = Tokenizer()
 
-    if should_train:
-        lines = load_kawa()
+    # if should_train:
+    lines = load_kawa()
 
-        corpus = [clean_text(x) for x in lines]
-        print("Corpus:", corpus[:2])
+    corpus = [clean_text(x) for x in lines]
+    print("Corpus:", corpus[:5])
 
-        inp_sequences, total_words = get_sequence_of_tokens(corpus, tokenizer)
-        predictors, label, max_sequence_len = generate_padded_sequences(inp_sequences, total_words)
-        model = create_model(max_sequence_len, total_words)
-        model.summary()
+    inp_sequences, total_words = get_sequence_of_tokens(corpus, tokenizer)
+    predictors, label, max_sequence_len = generate_padded_sequences(inp_sequences, total_words)
+    model = create_model(max_sequence_len, total_words)
+    model.summary()
 
-        model.fit(predictors, label, epochs=nb_epoch, verbose=5)
-        # model.save(model_file)
+    model.fit(predictors, label, epochs=nb_epoch, verbose=5)
+    # model.save(model_file)
     # else: # FIXME: Load and predict
-    #     model = load_model(model_file)
+    # model = load_model(model_file)
 
-    for sample in ["",
-                   "L'étoile du sol",
-                   "Elle me l'a toujours dit",
-                   "Les punchlines sont pour ceux"]:
-        nb_words = 200
+    for sample in load_seeds(lines):
         print(generate_text(model, tokenizer, sample, nb_words, max_sequence_len))
 
-    with open("../output/lstm.txt", "a") as f:
+    with open("./output/lstm.txt", "a+") as f:
         while True:
             input_text = input("> ")
             text = generate_text(model, tokenizer, input_text, nb_words, max_sequence_len)
diff --git a/KoozDawa/output/selection.txt b/KoozDawa/output/selection.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/KoozDawa/output/selection.txt