🤖 AI Model Lifecycle Guide

Example: Predict if an email is spam or not (binary classification).

Define input (email text) and output (spam / not spam).

import pandas as pd
data = pd.read_csv("emails.csv")

Steps may include:

• Tokenization (for text)
• Normalization (for numerical data)
• One-hot encoding (for categories)
• Splitting into train/test

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(data["text"], data["label"])

import torch.nn as nn

class SpamClassifier(nn.Module):
    def __init__(self, input_size):
        super().__init__()
        self.fc = nn.Linear(input_size, 1)

    def forward(self, x):
        return torch.sigmoid(self.fc(x))

# Define model, loss, optimizer
model = SpamClassifier(input_size=100)
criterion = nn.BCELoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

# Training loop
for epoch in range(10):
    optimizer.zero_grad()
    outputs = model(X_train_tensor)
    loss = criterion(outputs, y_train_tensor)
    loss.backward()
    optimizer.step()

from sklearn.metrics import accuracy_score

preds = model(X_test_tensor).round()
acc = accuracy_score(y_test_tensor, preds)
print("Accuracy:", acc)

torch.save(model.state_dict(), "spam_model.pt")

from fastapi import FastAPI
import torch

app = FastAPI()
model = SpamClassifier(input_size=100)
model.load_state_dict(torch.load("spam_model.pt"))
model.eval()

@app.post("/predict")
def predict(email_text: str):
    vector = preprocess(email_text)  # vectorize input
    with torch.no_grad():
        output = model(vector)
    return {"spam": bool(output.item() > 0.5)}

You might:

• Monitor prediction accuracy over time.
• Detect concept drift (data distribution changes).
• Retrain model with new data if needed.

As the model gets more data or starts performing poorly:

• Collect new labeled examples.
• Fine-tune or retrain the model.
• Repeat training and deployment.