File size: 5,334 Bytes
7812756 |
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 |
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Dict, Tuple
class KickstarterModel(nn.Module):
"""Kickstarter Project Success Prediction Model"""
def __init__(
self,
desc_embedding_dim=768,
blurb_embedding_dim=384,
risk_embedding_dim=384,
subcategory_embedding_dim=100,
category_embedding_dim=15,
country_embedding_dim=100,
numerical_features_dim=9,
hidden_dim=512,
dropout_rate=0.3
):
"""
Initialize the model
Args:
desc_embedding_dim: Description embedding vector dimension
blurb_embedding_dim: Blurb embedding vector dimension
risk_embedding_dim: Risk embedding vector dimension
subcategory_embedding_dim: Subcategory embedding vector dimension
category_embedding_dim: Category embedding vector dimension
country_embedding_dim: Country embedding vector dimension
numerical_features_dim: Numerical features dimension
hidden_dim: Hidden layer dimension
dropout_rate: Dropout rate
"""
super(KickstarterModel, self).__init__()
# Helper function to create feature processing layers
def create_fc_block(input_dim, output_dim):
return nn.Sequential(
nn.Linear(input_dim, output_dim),
nn.BatchNorm1d(output_dim),
nn.ReLU(),
nn.Dropout(dropout_rate)
)
# Feature processing layers
self.desc_fc = create_fc_block(desc_embedding_dim, hidden_dim)
self.blurb_fc = create_fc_block(blurb_embedding_dim, hidden_dim // 2)
self.risk_fc = create_fc_block(risk_embedding_dim, hidden_dim // 2)
self.subcategory_fc = create_fc_block(subcategory_embedding_dim, hidden_dim // 4)
self.category_fc = create_fc_block(category_embedding_dim, hidden_dim // 8)
self.country_fc = create_fc_block(country_embedding_dim, hidden_dim // 8)
self.numerical_fc = create_fc_block(numerical_features_dim, hidden_dim // 4)
# Combined features dimension
concat_dim = (hidden_dim +
hidden_dim // 2 +
hidden_dim // 2 +
hidden_dim // 4 +
hidden_dim // 8 +
hidden_dim // 8 +
hidden_dim // 4)
# Fully connected layers
self.fc1 = create_fc_block(concat_dim, hidden_dim)
self.fc2 = create_fc_block(hidden_dim, hidden_dim // 2)
# Output layer
self.output = nn.Linear(hidden_dim // 2, 1)
# Input names for SHAP explanation
self.input_names = [
'description_embedding',
'blurb_embedding',
'risk_embedding',
'subcategory_embedding',
'category_embedding',
'country_embedding',
'numerical_features'
]
def forward(self, inputs: Dict[str, torch.Tensor]) -> Tuple[torch.Tensor, Dict[str, torch.Tensor]]:
"""
Forward propagation
Args:
inputs: Dictionary containing all input features
Returns:
Prediction probability and intermediate feature representations
"""
# Process embeddings
desc_out = self.desc_fc(inputs['description_embedding'])
blurb_out = self.blurb_fc(inputs['blurb_embedding'])
risk_out = self.risk_fc(inputs['risk_embedding'])
subcategory_out = self.subcategory_fc(inputs['subcategory_embedding'])
category_out = self.category_fc(inputs['category_embedding'])
country_out = self.country_fc(inputs['country_embedding'])
numerical_out = self.numerical_fc(inputs['numerical_features'])
# Concatenate all features
combined = torch.cat([
desc_out,
blurb_out,
risk_out,
subcategory_out,
category_out,
country_out,
numerical_out
], dim=1)
# Fully connected layers
x = self.fc1(combined)
x = self.fc2(x)
# Output layer
logits = self.output(x)
probs = torch.sigmoid(logits)
# Store intermediate features for SHAP explanation
intermediate_features = {
'description_embedding': desc_out,
'blurb_embedding': blurb_out,
'risk_embedding': risk_out,
'subcategory_embedding': subcategory_out,
'category_embedding': category_out,
'country_embedding': country_out,
'numerical_features': numerical_out,
'combined': combined,
'fc1': x
}
return probs.squeeze(1), intermediate_features
def predict(self, inputs: Dict[str, torch.Tensor]) -> torch.Tensor:
"""
Prediction function
Args:
inputs: Dictionary containing all input features
Returns:
Prediction probability
"""
self.eval()
with torch.no_grad():
probs, _ = self.forward(inputs)
return probs |