diff --git a/8382/Ershov/lb/6/main.py b/8382/Ershov/lb/6/main.py
new file mode 100644
index 00000000..9d53b13b
--- /dev/null
+++ b/8382/Ershov/lb/6/main.py
@@ -0,0 +1,100 @@
+import string
+import matplotlib.pyplot as plt
+import numpy as np
+from keras import models
+from keras import layers
+from keras.datasets import imdb
+from tensorflow.keras import datasets
+
+
+def vectorize(sequences, dimension=10000):
+    results = np.zeros((len(sequences), dimension))
+    for i, sequence in enumerate(sequences):
+        results[i, sequence] = 1
+    return results
+
+
+def prepInputText(text, target):
+    def genNum(data, dic):
+        data = data.translate(str.maketrans(dict.fromkeys(string.punctuation))).split()
+        for i in range(len(data)):
+            num = dic.get(data[i])
+            if (num == None):
+                data[i] = 0
+            else:
+                data[i] = num
+        return data
+
+    dic = dict(datasets.imdb.get_word_index())
+    test_x = []
+    test_y = np.array(target).astype("float32")
+    for i in range(0, len(text)):
+        test_x.append(genNum(text[i], dic))
+    test_x = vectorize(test_x)
+    return test_x, test_y
+
+
+x = [
+    "This film is very interesting",
+    "Actors played great in this film",
+    "Uninteresting film",
+    "Cracking. Keeps attention from start to finish",
+    "In my opinion the movie is bad"
+]
+y = [1, 1, 0, 1, 0]
+
+
+dimension = 10000
+
+(training_data, training_targets), (testing_data, testing_targets) = imdb.load_data(num_words=dimension)
+data = np.concatenate((training_data, testing_data), axis=0)
+targets = np.concatenate((training_targets, testing_targets), axis=0)
+data = vectorize(data, dimension)
+targets = np.array(targets).astype("float32")
+
+test_x = data[:10000]
+test_y = targets[:10000]
+train_x = data[10000:]
+train_y = targets[10000:]
+
+model = models.Sequential()
+
+model.add(layers.Dense(50, activation="relu", input_shape=(dimension,)))
+model.add(layers.Dropout(0.4, noise_shape=None, seed=None))
+model.add(layers.Dense(30, activation="relu"))
+model.add(layers.Dropout(0.4, noise_shape=None, seed=None))
+model.add(layers.Dense(30, activation="relu"))
+model.add(layers.Dense(1, activation="sigmoid"))
+
+model.compile(optimizer="adam", loss="binary_crossentropy", metrics=["accuracy"])
+h = model.fit(train_x, train_y, epochs=2, batch_size=500, validation_data=(test_x, test_y))
+
+test_x, test_y = prepInputText(x, y)
+custom_loss, custom_acc = model.evaluate(test_x, test_y)
+print('custom_acc:', custom_acc)
+preds = model.predict(test_x)
+plt.figure(3, figsize=(8,5))
+plt.title("Custom dataset predications")
+plt.plot(test_y, 'r', marker='v', label='truth')
+plt.plot(preds, 'b', marker='x', label='pred')
+plt.legend()
+plt.show()
+plt.clf()
+
+
+plt.figure(1, figsize=(8, 5))
+plt.title("Training and test accuracy")
+plt.plot(h.history['acc'], 'r', label='train')
+plt.plot(h.history['val_acc'], 'b', label='test')
+plt.legend()
+plt.show()
+plt.clf()
+
+
+plt.figure(1, figsize=(8, 5))
+plt.title("Training and test loss")
+plt.plot(h.history['loss'], 'r', label='train')
+plt.plot(h.history['val_loss'], 'b', label='test')
+plt.legend()
+plt.show()
+plt.clf()
\ No newline at end of file
diff --git a/8382/Ershov/lb/6/report.pdf b/8382/Ershov/lb/6/report.pdf
new file mode 100644
index 00000000..b3ebecff
Binary files /dev/null and b/8382/Ershov/lb/6/report.pdf differ