File size: 11,860 Bytes
3bce187
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
"""
Base Sequential Model Architecture
Common interface and utilities for RNN, GRU, LSTM, and Transformer models.
"""

from abc import ABC, abstractmethod
from typing import Any, Dict, Tuple
import math

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import Tensor


class BaseSequentialModel(ABC, nn.Module):
    """Abstract base class for sequential classifiers."""

    def __init__(
        self,
        input_size: int = 512,
        hidden_size: int = 256,
        num_layers: int = 2,
        output_size: int = 43,
        dropout: float = 0.3,
        device: str = "cuda",
    ):
        super().__init__()
        self.input_size = input_size
        self.hidden_size = hidden_size
        self.num_layers = num_layers
        self.output_size = output_size
        self.dropout = dropout
        self.device = device

    @abstractmethod
    def forward(self, x: Tensor) -> Tensor:
        pass

    @abstractmethod
    def get_model_name(self) -> str:
        pass

    @abstractmethod
    def get_config_dict(self) -> Dict[str, Any]:
        pass

    def get_num_parameters(self) -> int:
        return sum(p.numel() for p in self.parameters() if p.requires_grad)

    def get_trainable_parameters(self) -> int:
        return self.get_num_parameters()


class RNNModel(BaseSequentialModel):
    """Vanilla RNN model for sequence classification."""

    def __init__(
        self,
        input_size: int = 512,
        hidden_size: int = 256,
        num_layers: int = 2,
        output_size: int = 43,
        dropout: float = 0.3,
        bidirectional: bool = True,
        device: str = "cuda",
    ):
        super().__init__(input_size, hidden_size, num_layers, output_size, dropout, device)
        self.bidirectional = bidirectional
        self.rnn = nn.RNN(
            input_size=input_size,
            hidden_size=hidden_size,
            num_layers=num_layers,
            batch_first=True,
            dropout=dropout if num_layers > 1 else 0.0,
            bidirectional=bidirectional,
        )
        out_size = hidden_size * (2 if bidirectional else 1)
        self.fc1 = nn.Linear(out_size, hidden_size)
        self.dropout_layer = nn.Dropout(dropout)
        self.fc2 = nn.Linear(hidden_size, output_size)

    def forward(self, x: Tensor) -> Tensor:
        rnn_out, _ = self.rnn(x)
        pooled = rnn_out.mean(dim=1)
        hidden = F.relu(self.fc1(pooled))
        hidden = self.dropout_layer(hidden)
        return self.fc2(hidden)

    def get_model_name(self) -> str:
        return "RNN"

    def get_config_dict(self) -> Dict[str, Any]:
        return {
            "model": "RNN",
            "input_size": self.input_size,
            "hidden_size": self.hidden_size,
            "num_layers": self.num_layers,
            "output_size": self.output_size,
            "dropout": self.dropout,
            "bidirectional": self.bidirectional,
            "num_parameters": self.get_num_parameters(),
        }


class GRUModel(BaseSequentialModel):
    """GRU model for sequence classification."""

    def __init__(
        self,
        input_size: int = 512,
        hidden_size: int = 256,
        num_layers: int = 2,
        output_size: int = 43,
        dropout: float = 0.3,
        bidirectional: bool = True,
        device: str = "cuda",
    ):
        super().__init__(input_size, hidden_size, num_layers, output_size, dropout, device)
        self.bidirectional = bidirectional
        self.gru = nn.GRU(
            input_size=input_size,
            hidden_size=hidden_size,
            num_layers=num_layers,
            batch_first=True,
            dropout=dropout if num_layers > 1 else 0.0,
            bidirectional=bidirectional,
        )
        out_size = hidden_size * (2 if bidirectional else 1)
        self.fc1 = nn.Linear(out_size, hidden_size)
        self.dropout_layer = nn.Dropout(dropout)
        self.fc2 = nn.Linear(hidden_size, output_size)

    def forward(self, x: Tensor) -> Tensor:
        gru_out, _ = self.gru(x)
        pooled = gru_out.mean(dim=1)
        hidden = F.relu(self.fc1(pooled))
        hidden = self.dropout_layer(hidden)
        return self.fc2(hidden)

    def get_model_name(self) -> str:
        return "GRU"

    def get_config_dict(self) -> Dict[str, Any]:
        return {
            "model": "GRU",
            "input_size": self.input_size,
            "hidden_size": self.hidden_size,
            "num_layers": self.num_layers,
            "output_size": self.output_size,
            "dropout": self.dropout,
            "bidirectional": self.bidirectional,
            "num_parameters": self.get_num_parameters(),
        }


class AdditiveAttention(nn.Module):
    """Additive attention for sequence pooling."""

    def __init__(self, hidden_size: int):
        super().__init__()
        self.score = nn.Sequential(
            nn.Linear(hidden_size, hidden_size),
            nn.Tanh(),
            nn.Linear(hidden_size, 1),
        )

    def forward(self, sequence_outputs: Tensor) -> Tuple[Tensor, Tensor]:
        weights = torch.softmax(self.score(sequence_outputs), dim=1)
        context = torch.sum(weights * sequence_outputs, dim=1)
        return context, weights


class SinusoidalPositionalEncoding(nn.Module):
    """Sinusoidal positional encoding."""

    def __init__(self, d_model: int, max_len: int = 512, dropout: float = 0.1):
        super().__init__()
        self.dropout = nn.Dropout(dropout)

        position = torch.arange(0, max_len, dtype=torch.float32).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2, dtype=torch.float32) * (-math.log(10000.0) / d_model))
        pe = torch.zeros(max_len, d_model, dtype=torch.float32)
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        self.register_buffer("pe", pe.unsqueeze(0), persistent=False)

    def forward(self, x: Tensor) -> Tensor:
        seq_len = x.size(1)
        x = x + self.pe[:, :seq_len]
        return self.dropout(x)


class LSTMModel(BaseSequentialModel):
    """LSTM model with additive attention."""

    def __init__(
        self,
        input_size: int = 512,
        hidden_size: int = 256,
        num_layers: int = 2,
        output_size: int = 43,
        dropout: float = 0.3,
        bidirectional: bool = True,
        device: str = "cuda",
    ):
        super().__init__(input_size, hidden_size, num_layers, output_size, dropout, device)
        self.bidirectional = bidirectional
        self.lstm = nn.LSTM(
            input_size=input_size,
            hidden_size=hidden_size,
            num_layers=num_layers,
            batch_first=True,
            dropout=dropout if num_layers > 1 else 0.0,
            bidirectional=bidirectional,
        )
        out_size = hidden_size * (2 if bidirectional else 1)
        self.attention = AdditiveAttention(out_size)
        self.fc1 = nn.Linear(out_size, hidden_size)
        self.dropout_layer = nn.Dropout(dropout)
        self.fc2 = nn.Linear(hidden_size, output_size)

    def forward(self, x: Tensor) -> Tensor:
        lstm_out, _ = self.lstm(x)
        context, _ = self.attention(lstm_out)
        hidden = F.relu(self.fc1(context))
        hidden = self.dropout_layer(hidden)
        return self.fc2(hidden)

    def get_model_name(self) -> str:
        return "LSTM"

    def get_config_dict(self) -> Dict[str, Any]:
        return {
            "model": "LSTM",
            "input_size": self.input_size,
            "hidden_size": self.hidden_size,
            "num_layers": self.num_layers,
            "output_size": self.output_size,
            "dropout": self.dropout,
            "bidirectional": self.bidirectional,
            "num_parameters": self.get_num_parameters(),
        }


class TransformerModel(BaseSequentialModel):
    """Transformer encoder model for sequence classification - optimized for CPU."""

    def __init__(
        self,
        input_size: int = 512,
        hidden_size: int = 128,  # Reduced from 256
        num_layers: int = 1,  # Reduced from 2
        output_size: int = 43,
        dropout: float = 0.2,  # Reduced from 0.3
        bidirectional: bool = True,
        device: str = "cuda",
        attention_heads: int = 4,  # Reduced from 8
        ffn_dim: int = 256,  # Reduced from 1024
        max_seq_len: int = 512,
    ):
        super().__init__(input_size, hidden_size, num_layers, output_size, dropout, device)
        self.bidirectional = bidirectional
        self.attention_heads = attention_heads
        self.ffn_dim = ffn_dim
        self.max_seq_len = max_seq_len

        self.input_projection = nn.Linear(input_size, hidden_size)
        self.positional_encoding = SinusoidalPositionalEncoding(hidden_size, max_len=max_seq_len, dropout=dropout)
        encoder_layer = nn.TransformerEncoderLayer(
            d_model=hidden_size,
            nhead=attention_heads,
            dim_feedforward=ffn_dim,
            dropout=dropout,
            batch_first=True,
            activation="relu",  # Changed from gelu to relu for faster computation
        )
        self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)

        self.fc1 = nn.Linear(hidden_size, hidden_size)
        self.dropout_layer = nn.Dropout(dropout)
        self.fc2 = nn.Linear(hidden_size, output_size)

    def forward(self, x: Tensor) -> Tensor:
        if x.size(1) > self.max_seq_len:
            raise ValueError(f"Sequence length {x.size(1)} exceeds max_seq_len={self.max_seq_len}")
        x = self.input_projection(x)
        x = self.positional_encoding(x)
        x = self.transformer(x)
        pooled = x.mean(dim=1)
        hidden = F.relu(self.fc1(pooled))
        hidden = self.dropout_layer(hidden)
        return self.fc2(hidden)

    def get_model_name(self) -> str:
        return "Transformer"

    def get_config_dict(self) -> Dict[str, Any]:
        return {
            "model": "Transformer",
            "input_size": self.input_size,
            "hidden_size": self.hidden_size,
            "num_layers": self.num_layers,
            "output_size": self.output_size,
            "dropout": self.dropout,
            "attention_heads": self.attention_heads,
            "ffn_dim": self.ffn_dim,
            "max_seq_len": self.max_seq_len,
            "num_parameters": self.get_num_parameters(),
        }


def create_model(model_name: str, config: Dict[str, Any], device: str = "cuda") -> BaseSequentialModel:
    model_map = {
        "rnn": RNNModel,
        "gru": GRUModel,
        "lstm": LSTMModel,
        "transformer": TransformerModel,
    }
    if model_name.lower() not in model_map:
        raise ValueError(f"Unknown model: {model_name}. Choose from {list(model_map.keys())}")
    model_class = model_map[model_name.lower()]
    return model_class(**config, device=device)


def get_model_summary(model: BaseSequentialModel) -> Dict[str, Any]:
    return {
        "name": model.get_model_name(),
        "config": model.get_config_dict(),
        "total_params": model.get_num_parameters(),
        "trainable_params": model.get_trainable_parameters(),
    }


if __name__ == "__main__":
    device = "cuda" if torch.cuda.is_available() else "cpu"
    x = torch.randn(4, 10, 512).to(device)

    rnn_model = RNNModel(input_size=512, hidden_size=256, num_layers=2, output_size=43, device=device).to(device)
    rnn_out = rnn_model(x)
    print(f"RNN output shape: {rnn_out.shape}")

    gru_model = GRUModel(input_size=512, hidden_size=256, num_layers=2, output_size=43, device=device).to(device)
    gru_out = gru_model(x)
    print(f"GRU output shape: {gru_out.shape}")

    print("\n✓ Base sequential models tests passed!")


# Backward-compatible aliases for older imports in notebooks/scripts.
RNN_Model = RNNModel
GRU_Model = GRUModel