Edit app.py

app.py

@@ -1,5 +1,100 @@
+import os
+import numpy as np
+import tensorflow as tf
+import pandas as pd
+import time
+
+from recommenders.models.sasrec.model import SASREC
+from tabulate import tabulate
+
 import streamlit as st
 
-x = st.slider('Select a value')
-st.write(x, 'squared is', x * x)
 
+class SASREC_Vessl(SASREC):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+    def predict_next(self, input):
+        # seq generation
+        training = False
+        seq = np.zeros([self.seq_max_len], dtype=np.int32)
+        idx = self.seq_max_len - 1
+        idx -= 1
+        for i in input[::-1]:
+            seq[idx] = i
+            idx -= 1
+            if idx == -1:
+                break
+
+        input_seq = np.array([seq])
+        candidate = np.expand_dims(np.arange(1, self.item_num + 1, 1), axis=0)
+
+        mask = tf.expand_dims(tf.cast(tf.not_equal(input_seq, 0), tf.float32),
+                              -1)
+        seq_embeddings, positional_embeddings = self.embedding(input_seq)
+        seq_embeddings += positional_embeddings
+        seq_embeddings *= mask
+        seq_attention = seq_embeddings
+        seq_attention = self.encoder(seq_attention, training, mask)
+        seq_attention = self.layer_normalization(seq_attention)  # (b, s, d)
+        seq_emb = tf.reshape(
+            seq_attention,
+            [tf.shape(input_seq)[0] * self.seq_max_len, self.embedding_dim],
+        )  # (b*s, d)
+        candidate_emb = self.item_embedding_layer(candidate)  # (b, s, d)
+        candidate_emb = tf.transpose(candidate_emb, perm=[0, 2, 1])  # (b, d, s)
+
+        test_logits = tf.matmul(seq_emb, candidate_emb)
+        test_logits = tf.reshape(
+            test_logits,
+            [tf.shape(input_seq)[0], self.seq_max_len, self.item_num],
+        )
+        test_logits = test_logits[:, -1, :]  # (1, 101)
+
+        predictions = np.array(test_logits)[0]
+
+        return predictions
+
+
+def load_model():
+    model_config = {
+        "MAXLEN": 50,
+        "NUM_BLOCKS": 2,  # NUMBER OF TRANSFORMER BLOCKS
+        "HIDDEN_UNITS": 100,  # NUMBER OF UNITS IN THE ATTENTION CALCULATION
+        "NUM_HEADS": 1,  # NUMBER OF ATTENTION HEADS
+        "DROPOUT_RATE": 0.2,  # DROPOUT RATE
+        "L2_EMB": 0.0,  # L2 REGULARIZATION COEFFICIENT
+        "NUM_NEG_TEST": 100,
+        # NUMBER OF NEGATIVE EXAMPLES PER POSITIVE EXAMPLE
+    }
+
+    model = SASREC_Vessl(
+        item_num=12101,  # should be changed according to dataset
+        seq_max_len=model_config.get("MAXLEN"),
+        num_blocks=model_config.get("NUM_BLOCKS"),
+        embedding_dim=model_config.get("HIDDEN_UNITS"),
+        attention_dim=model_config.get("HIDDEN_UNITS"),
+        attention_num_heads=model_config.get("NUM_HEADS"),
+        dropout_rate=model_config.get("DROPOUT_RATE"),
+        conv_dims=[100, 100],
+        l2_reg=model_config.get("L2_EMB"),
+        num_neg_test=model_config.get("NUM_NEG_TEST"),
+    )
+
+    if os.path.isfile('best.index') and os.path.isfile('best.data-00000-of-00001'):
+        model.load_weights('best')
+
+    return model
+
+
+def main():
+    st.title('Self-Attentive Sequential Recommendation(SASRec)')
+    model = load_model()
+    st.write(model)
+
+    numbers = st.text_input
+    st.write(numbers)
+
+
+if __name__ == '__main__':
+    main()