ml-expert — скилл для Claude Code: что делает и как установить

---
name: ml-expert
version: 1.1.0
description: >-
  Expert-level machine learning, deep learning, model training, and MLOps. Use when the
  user mentions machine learning, deep learning, neural networks, MLOps, or data science,
  or when the task involves Machine Learning Fundamentals, Data Preparation, or Model
  Training.
category: ai
tags: [machine-learning, deep-learning, neural-networks, mlops, data-science]
allowed-tools:
  - Read
  - Write
  - Edit
  - Bash(python:*)
---

# Machine Learning Expert

Expert guidance for machine learning systems, deep learning, model training, deployment, and MLOps practices.

## Core Concepts

### Machine Learning Fundamentals

- Supervised learning (classification, regression)
- Unsupervised learning (clustering, dimensionality reduction)
- Reinforcement learning
- Feature engineering
- Model evaluation and validation
- Hyperparameter tuning

### Deep Learning

- Neural networks (CNNs, RNNs, Transformers)
- Transfer learning
- Fine-tuning pre-trained models
- Attention mechanisms
- GANs (Generative Adversarial Networks)
- Autoencoders

### MLOps

- Model versioning and tracking
- Experiment management
- Model deployment and serving
- Monitoring and retraining
- CI/CD for ML pipelines
- A/B testing for models

## Supervised Learning

```python
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, confusion_matrix
import joblib

class MLPipeline:
    def __init__(self):
        self.scaler = StandardScaler()
        self.model = None
        self.feature_names = None

    def prepare_data(self, X: pd.DataFrame, y: pd.Series, test_size: float = 0.2):
        """Split and scale data"""
        X_train, X_test, y_train, y_test = train_test_split(
            X, y, test_size=test_size, random_state=42, stratify=y
        )

        # Scale features
        X_train_scaled = self.scaler.fit_transform(X_train)
        X_test_scaled = self.scaler.transform(X_test)

        self.feature_names = X.columns.tolist()

        return X_train_scaled, X_test_scaled, y_train, y_test

    def train_classifier(self, X_train, y_train, n_estimators: int = 100):
        """Train random forest classifier"""
        self.model = RandomForestClassifier(
            n_estimators=n_estimators,
            max_depth=10,
            random_state=42,
            n_jobs=-1
        )

        self.model.fit(X_train, y_train)

        # Cross-validation
        cv_scores = cross_val_score(self.model, X_train, y_train, cv=5)

        return {
            "cv_mean": cv_scores.mean(),
            "cv_std": cv_scores.std(),
            "feature_importance": dict(zip(
                self.feature_names,
                self.model.feature_importances_
            ))
        }

    def evaluate(self, X_test, y_test) -> dict:
        """Evaluate model performance"""
        y_pred = self.model.predict(X_test)
        y_proba = self.model.predict_proba(X_test)

        return {
            "predictions": y_pred,
            "probabilities": y_proba,
            "confusion_matrix": confusion_matrix(y_test, y_pred).tolist(),
            "classification_report": classification_report(y_test, y_pred, output_dict=True)
        }

    def save_model(self, path: str):
        """Save model and scaler"""
        joblib.dump({
            "model": self.model,
            "scaler": self.scaler,
            "feature_names": self.feature_names
        }, path)
```

## Deep Learning with PyTorch

```python
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, Dataset

class NeuralNetwork(nn.Module):
    def __init__(self, input_size: int, hidden_size: int, num_classes: int):
        super().__init__()
        self.fc1 = nn.Linear(input_size, hidden_size)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.3)
        self.fc2 = nn.Linear(hidden_size, hidden_size // 2)
        self.fc3 = nn.Linear(hidden_size // 2, num_classes)

    def forward(self, x):
        x = self.fc1(x)
        x = self.relu(x)
        x = self.dropout(x)
        x = self.fc2(x)
        x = self.relu(x)
        x = self.fc3(x)
        return x

class Trainer:
    def __init__(self, model, device='cuda' if torch.cuda.is_available() else 'cpu'):
        self.model = model.to(device)
        self.device = device
        self.criterion = nn.CrossEntropyLoss()
        self.optimizer = optim.Adam(model.parameters(), lr=0.001)

    def train_epoch(self, dataloader: DataLoader) -> float:
        """Train for one epoch"""
        self.model.train()
        total_loss = 0

        for batch_idx, (data, target) in enumerate(dataloader):
            data, target = data.to(self.device), target.to(self.device)

            self.optimizer.zero_grad()
            output = self.model(data)
            loss = self.criterion(output, target)

            loss.backward()
            self.optimizer.step()

            total_loss += loss.item()

        return total_loss / len(dataloader)

    def evaluate(self, dataloader: DataLoader) -> dict:
        """Evaluate model"""
        self.model.eval()
        correct = 0
        total = 0

        with torch.no_grad():
            for data, target in dataloader:
                data, target = data.to(self.device), target.to(self.device)
                output = self.model(data)
                _, predicted = torch.max(output.data, 1)
                total += target.size(0)
                correct += (predicted == target).sum().item()

        return {
            "accuracy": 100 * correct / total,
            "total_samples": total
        }

    def train(self, train_loader: DataLoader, val_loader: DataLoader,
              epochs: int = 10):
        """Full training loop"""
        history = {"train_loss": [], "val_acc": []}

        for epoch in range(epochs):
            train_loss = self.train_epoch(train_loader)
            val_metrics = self.evaluate(val_loader)

            history["train_loss"].append(train_loss)
            history["val_acc"].append(val_metrics["accuracy"])

            print(f"Epoch {epoch+1}/{epochs} - Loss: {train_loss:.4f} - Val Acc: {val_metrics['accuracy']:.2f}%")

        return history
```

## Model Deployment

```python
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import numpy as np

app = FastAPI()

class PredictionRequest(BaseModel):
    features: list[float]

class PredictionResponse(BaseModel):
    prediction: int
    probability: float
    model_version: str

class ModelServer:
    def __init__(self, model_path: str):
        self.model_data = joblib.load(model_path)
        self.model = self.model_data["model"]
        self.scaler = self.model_data["scaler"]
        self.version = "1.0.0"

    def predict(self, features: np.ndarray) -> dict:
        """Make prediction"""
        # Scale features
        features_scaled = self.scaler.transform(features.reshape(1, -1))

        # Predict
        prediction = self.model.predict(features_scaled)[0]
        probability = self.model.predict_proba(features_scaled)[0].max()

        return {
            "prediction": int(prediction),
            "probability": float(probability),
            "model_version": self.version
        }

# Global model instance
model_server = ModelServer("model.pkl")

@app.post("/predict", response_model=PredictionResponse)
async def predict(request: PredictionRequest):
    try:
        features = np.array(request.features)
        result = model_server.predict(features)
        return PredictionResponse(**result)
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/health")
async def health():
    return {"status": "healthy", "model_version": model_server.version}
```

## MLOps with MLflow

```python
import mlflow
import mlflow.sklearn
from mlflow.tracking import MlflowClient

class MLflowExperiment:
    def __init__(self, experiment_name: str):
        mlflow.set_experiment(experiment_name)
        self.client = MlflowClient()

    def log_training_run(self, model, X_train, y_train, X_test, y_test,
                        params: dict):
        """Log training run with MLflow"""
        with mlflow.start_run():
            # Log parameters
            mlflow.log_params(params)

            # Train model
            model.fit(X_train, y_train)

            # Evaluate
            train_score = model.score(X_train, y_train)
            test_score = model.score(X_test, y_test)

            # Log metrics
            mlflow.log_metric("train_accuracy", train_score)
            mlflow.log_metric("test_accuracy", test_score)

            # Log model
            mlflow.sklearn.log_model(model, "model")

            # Log feature importance
            if hasattr(model, 'feature_importances_'):
                feature_importance = dict(enumerate(model.feature_importances_))
                mlflow.log_dict(feature_importance, "feature_importance.json")

            run_id = mlflow.active_run().info.run_id
            return run_id

    def register_model(self, run_id: str, model_name: str):
        """Register model in MLflow model registry"""
        model_uri = f"runs:/{run_id}/model"
        mlflow.register_model(model_uri, model_name)

    def promote_to_production(self, model_name: str, version: int):
        """Promote model version to production"""
        self.client.transition_model_version_stage(
            name=model_name,
            version=version,
            stage="Production"
        )
```

## Best Practices

### Data Preparation

- Handle missing values appropriately
- Scale/normalize features
- Encode categorical variables properly
- Split data before any preprocessing
- Use stratified splits for imbalanced data
- Create validation set for hyperparameter tuning

### Model Training

- Start with simple baselines
- Use cross-validation
- Monitor training and validation metrics
- Implement early stopping
- Save best model checkpoints
- Track experiments systematically

### Deployment

- Version models and datasets
- Monitor model performance in production
- Implement model A/B testing
- Set up retraining pipelines
- Log predictions for analysis
- Implement fallback mechanisms

## Anti-Patterns

❌ Training on test data (data leakage)
❌ No validation set for hyperparameter tuning
❌ Ignoring class imbalance
❌ Not scaling features
❌ Overfitting to training data
❌ No model versioning
❌ Missing monitoring in production

## Resources

- Scikit-learn: https://scikit-learn.org/
- PyTorch: https://pytorch.org/
- TensorFlow: https://www.tensorflow.org/
- MLflow: https://mlflow.org/
- Hugging Face: https://huggingface.co/