Streamlit App

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+media/background.jpg filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,99 @@
+import torch
+from tokenizers import Tokenizer
+from torch.utils.data import DataLoader
+
+import streamlit as st
+import base64
+from model import CustomDataset, TransformerEncoder
+
+st.set_page_config(layout="wide",page_title="TeknoFest We Bears NLP Competition", page_icon="./media/3bears.ico")
+
+tag2id = {"O": 0, "olumsuz": 1, "nötr": 2, "olumlu": 3, "org": 4}
+id2tag = {value: key for key, value in tag2id.items()}
+device = torch.device('cpu')
+
+@st.cache_resource
+def load_model_to_cpu(_model, path="model.pth"):
+    checkpoint = torch.load(path, map_location=torch.device('cpu'))
+    _model.load_state_dict(checkpoint)
+    return _model
+    
+def get_base64(bin_file):
+    with open(bin_file, 'rb') as f:
+        data = f.read()
+    return base64.b64encode(data).decode()
+
+def predict_fonk(model, device, example, tokenizer):
+    model.to(device)
+    model.eval()
+    predictions = []
+
+    encodings_prdict = tokenizer.encode(example)
+
+    predict_texts = [encodings_prdict.tokens]
+    predict_input_ids = [encodings_prdict.ids]
+    predict_attention_masks = [encodings_prdict.attention_mask]
+    predict_token_type_ids = [encodings_prdict.type_ids]
+    prediction_labels = [encodings_prdict.type_ids]
+
+    predict_data = CustomDataset(predict_texts, predict_input_ids, predict_attention_masks, predict_token_type_ids,
+                                 prediction_labels)
+
+    predict_loader = DataLoader(predict_data, batch_size=1, shuffle=False)
+
+    with torch.no_grad():
+        for dataset in predict_loader:
+            batch_input_ids = dataset['input_ids'].to(device)
+            batch_att_mask = dataset['attention_mask'].to(device)
+
+
+
+            outputs = model(batch_input_ids, batch_att_mask)
+            logits = outputs.view(-1, outputs.size(-1))  # Flatten the outputs
+            _, predicted = torch.max(logits, 1)
+
+            # Ignore padding tokens for predictions
+            predictions.append(predicted)
+
+    results_list = []
+    entity_list = []
+    results_dict = {}
+    trio = zip(predict_loader.dataset[0]["text"], predictions[0].tolist(), predict_attention_masks[0])
+    
+    for i, (token, label, attention) in enumerate(trio):
+        if attention != 0 and label != 0 and label !=4:
+            for next_ones in predictions[0].tolist()[i+1:]:
+                i+=1
+                if next_ones == 4:
+                    token = token +" "+ predict_loader.dataset[0]["text"][i]
+                else:break
+            if token not in entity_list:
+                entity_list.append(token)
+                results_list.append({"entity":token,"sentiment":id2tag.get(label)})
+
+
+    results_dict["entity_list"] = entity_list
+    results_dict["results"] = results_list
+
+
+    return results_dict
+
+model = TransformerEncoder()
+model = load_model_to_cpu(model, "model.pth")
+tokenizer = Tokenizer.from_file("tokenizer.json")
+
+background = get_base64("./media/background.jpg")
+
+with open("./style/style.css", "r") as style:
+    css=f"""<style>{style.read().format(background=background)}</style>"""
+    st.markdown(css, unsafe_allow_html=True)
+
+left, middle, right = st.columns([1,1.5,1])
+main, comps , result = middle.tabs([" ", " ", " "])
+with main:
+    example = st.text_area(label='Metin Kutusu: ', placeholder="Lütfen Şikayet veya Yorum Metnini Buraya Yazın, daha sonra Predicte tıklayın")
+
+    if st.button("Predict"):
+        predict_list = predict_fonk(model=model, device=device, example=example, tokenizer=tokenizer)
+           
+        st.write(predict_list)

model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:320bc3db757e8b07b86ae43a0a3ff8adca691db7e25359b1e31d999ef4906d65
+size 280754978

model.py

@@ -0,0 +1,146 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.utils.data import Dataset
+import math
+
+class CustomDataset(Dataset):
+    def __init__(self, texts, input_ids, attention_masks, token_type_ids, labels):
+        self.texts = texts
+        self.input_ids = input_ids
+        self.token_type_ids = token_type_ids
+        self.attention_masks = attention_masks
+        self.labels = labels
+
+
+    def __len__(self):
+        return len(self.texts)
+
+    def __getitem__(self, item ):
+        text = self.texts[item]
+        input_id = torch.LongTensor(self.input_ids[item])
+        token_type_id = torch.LongTensor(self.token_type_ids[item])
+        attention_mask = torch.LongTensor(self.attention_masks[item])
+        label = torch.LongTensor(self.labels[item])
+
+
+        return {
+            'text': text,
+            'input_ids': input_id,
+            'token_type_ids': token_type_id,
+            'attention_mask': attention_mask,
+            'labels': label,
+        }
+class FeedForwardSubLayer(nn.Module):
+    # Specify the two linear layers' input and output sizes
+    def __init__(self, d_model, d_ff):
+        super(FeedForwardSubLayer, self).__init__()
+        self.fc1 = nn.Linear(d_model, d_ff)
+        self.fc2 = nn.Linear(d_ff, d_model)
+        self.relu = nn.ReLU()
+
+	# Apply a forward pass
+    def forward(self, x):
+        return self.fc2(self.relu(self.fc1(x)))
+
+# Complete the initialization of elements in the encoder layer
+class EncoderLayer(nn.Module):
+    def __init__(self, d_model, num_heads, d_ff, dropout):
+        super(EncoderLayer, self).__init__()
+        self.self_attn = MultiHeadAttention(d_model, num_heads)
+        self.feed_forward = FeedForwardSubLayer(d_model, d_ff)
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x, mask):
+        attn_output = self.self_attn(x, x, x, mask)
+        x = self.norm1(x + self.dropout(attn_output))
+        ff_output = self.feed_forward(x)
+        return self.norm2(x + self.dropout(ff_output))
+
+class MultiHeadAttention(nn.Module):
+    def __init__(self, d_model, num_heads):
+        super(MultiHeadAttention, self).__init__()
+        # Set the number of attention heads
+        self.num_heads = num_heads
+        self.d_model = d_model
+        assert d_model % num_heads == 0 #dimension, headlere tam bölünüyormu kontrol et.
+        self.head_dim = d_model // num_heads
+        # Set up the linear transformations
+        self.query_linear = nn.Linear(d_model, d_model)
+        self.key_linear = nn.Linear(d_model, d_model)
+        self.value_linear = nn.Linear(d_model, d_model)
+        self.output_linear = nn.Linear(d_model, d_model)
+
+    def split_heads(self, x, batch_size):
+        # Split the sequence embeddings in x across the attention heads
+        x = x.view(batch_size, -1, self.num_heads, self.head_dim)
+        return x.permute(0, 2, 1, 3) #.contiguous().view(batch_size * self.num_heads, -1, self.head_dim)
+
+    def compute_attention(self, query, key, mask=None):
+        # Compute dot-product attention scores
+        scores = torch.matmul(query, key.permute(0,1,3,2))
+        mask = mask.unsqueeze(1).unsqueeze(1)
+
+
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, float("-1e20"))
+        # Normalize attention scores into attention weights
+        attention_weights = F.softmax(scores, dim=-1)
+        return attention_weights
+
+    def forward(self, query, key, value, mask=None):
+        batch_size = query.size(0)
+
+        query = self.split_heads(self.query_linear(query), batch_size)
+        key = self.split_heads(self.key_linear(key), batch_size)
+        value = self.split_heads(self.value_linear(value), batch_size)
+
+        attention_weights = self.compute_attention(query, key, mask)
+
+        # Multiply attention weights by values, concatenate and linearly project outputs
+        output = torch.matmul(attention_weights, value)
+        output = output.view(batch_size, self.num_heads, -1, self.head_dim).permute(0, 2, 1, 3).contiguous().view(
+            batch_size, -1, self.d_model)
+        return self.output_linear(output)
+
+class PositionalEncoder(nn.Module):
+    def __init__(self, d_model, max_length):
+        super(PositionalEncoder, self).__init__()
+        self.d_model = d_model
+        self.max_length = max_length
+
+        # Initialize the positional encoding matrix
+        pe = torch.zeros(max_length, d_model)
+        position = torch.arange(0, max_length, dtype=torch.float).unsqueeze(1)
+        div_term = torch.exp(torch.arange(0, d_model, 2, dtype=torch.float) * -(math.log(10000.0) / d_model))
+
+        # Calculate and assign position encodings to the matrix
+        pe[:, 0::2] = torch.sin(position * div_term)
+        pe[:, 1::2] = torch.cos(position * div_term)
+        pe = pe.unsqueeze(0)
+        self.register_buffer('pe', pe)
+
+    # Update the embeddings tensor adding the positional encodings
+    def forward(self, x):
+        x = x + self.pe[:, :x.size(1)]
+        return x
+
+class TransformerEncoder(nn.Module):
+    def __init__(self):
+        super(TransformerEncoder, self).__init__()
+        self.embedding = nn.Embedding(100000, 512)
+        self.positional_encoding = PositionalEncoder(512, 128)
+        # Define a stack of multiple encoder layers
+        self.layers = nn.ModuleList([EncoderLayer(512, 8, 2048, 0.1) for _ in range(6)])
+
+    # Complete the forward pass method
+    def forward(self, x, mask):
+        x = self.embedding(x)
+        x = self.positional_encoding(x)
+        for layer in self.layers:
+            x = layer(x, mask)
+        return x
+
+

requirements.txt

@@ -0,0 +1,3 @@
+torch==2.3.0
+tokenizers==0.13.3
+streamlit

style/style.css

@@ -0,0 +1,26 @@
+body {{
+    background-image: url("data:image/png;base64,{background}");
+    background-size: cover;
+    background-repeat: no-repeat;
+    background-attachment: fixed;
+    background-position: center;
+    margin: 0;
+    padding: 0;
+    height: 100vh;
+    }}
+
+    p {{
+    color: black;
+    font-family: "Google Sans",Roboto,Arial,sans-serif;
+    font-size: 20px;
+    
+    }}
+
+    header[data-testid = "stHeader"]{{background: rgba(255,255,255,0);}}
+    button[data-testid = "baseButton-secondary"]{{width: 100%;}}
+    div[data-testid = "stButton"] > button > div > p {{color: white; font-size: 15px;}}
+    div[data-testid = "stApp"]{{background: None; color: black;}}
+    div[id^=tabs-bui][id$=-tabpanel-0]{{padding: 20px; border-radius: 2rem; background: rgba(255,255,255,0.7);}}
+    div[class = "react-json-view"]{{padding: 20px; border-radius: 2rem; background: rgba(255,255,255,0.7);}}
+    div[data-testid = "stMarkdownContainer"] > p {{color: black; font-size: 15px; font-weight: bold;}}
+