trustformers 0.1.1

//! Advanced Pipeline Composition Example
#![allow(clippy::all)]
#![allow(unused_variables)]
//!
//! This example demonstrates advanced pipeline composition, chaining,
//! and multi-modal processing capabilities in TrustformeRS.

use fastrand;
use std::sync::Arc;
use trustformers::pipeline::{Pipeline, PipelineOutput};
use trustformers::{pipeline, PipelineChain, Result};

#[tokio::main]
async fn main() -> Result<()> {
    println!("🔗 TrustformeRS Advanced Pipeline Composition Examples\n");

    // Pipeline Chaining Example
    pipeline_chaining_example().await?;

    // Multi-Modal Pipeline Example
    multimodal_example().await?;

    // Ensemble Pipeline Example
    ensemble_example().await?;

    // Conversational Pipeline Example
    conversational_example().await?;

    // Custom Pipeline Composition
    custom_composition_example().await?;

    println!("\n✅ All advanced examples completed successfully!");
    Ok(())
}

/// Demonstrate pipeline chaining for complex workflows
async fn pipeline_chaining_example() -> Result<()> {
    println!("🔗 Pipeline Chaining Example");
    println!("============================");

    // Create individual pipelines
    let generator = pipeline("text-generation", Some("gpt2-medium"), None)?;
    let summarizer = pipeline("summarization", Some("facebook/bart-large-cnn"), None)?;
    let classifier = pipeline(
        "text-classification",
        Some("cardiffnlp/twitter-roberta-base-sentiment-latest"),
        None,
    )?;

    // Create a pipeline chain: Generate -> Summarize -> Classify
    let chained_pipeline = PipelineChain::from_pipelines(vec![generator, summarizer, classifier]);

    // Test the chained pipeline
    let input_prompt = "Write a story about artificial intelligence";
    println!("Input prompt: \"{}\"", input_prompt);

    println!("\nExecuting pipeline chain:");
    println!("1. Executing chained pipeline...");
    let result = chained_pipeline.__call__(input_prompt.to_string())?;
    println!("   Result: {:?}", result);

    // Note: In a real implementation, you'd need proper type conversions
    // between pipeline outputs and inputs

    Ok(())
}

/// Demonstrate multi-modal pipeline processing
async fn multimodal_example() -> Result<()> {
    println!("🎭 Multi-Modal Pipeline Example");
    println!("===============================");

    // Create a multi-modal pipeline configuration
    let multimodal_config = r#"
    {
        "vision_model": "clip-vit-base-patch32",
        "text_model": "distilbert-base-uncased",
        "fusion_strategy": "cross_attention",
        "output_dim": 512
    }
    "#;

    // In a real implementation, this would process both image and text
    println!("Multi-modal configuration:");
    println!("{}", multimodal_config);

    // Simulate multi-modal processing
    let sample_inputs = vec![
        ("A beautiful sunset over the ocean", "image_sunset.jpg"),
        ("A cat sitting on a chair", "image_cat.jpg"),
        ("Technology and innovation", "image_tech.jpg"),
    ];

    println!("\nProcessing multi-modal inputs:");
    for (text, image_path) in &sample_inputs {
        println!("  Text: \"{}\"", text);
        println!("  Image: {}", image_path);
        println!("  Fused representation: [simulated 512-dim vector]");
        println!("  Similarity score: {:.3}", fastrand::f32() * 0.5 + 0.5);
        println!();
    }

    Ok(())
}

/// Demonstrate ensemble pipeline for improved accuracy
async fn ensemble_example() -> Result<()> {
    println!("🎯 Ensemble Pipeline Example");
    println!("============================");

    // Create multiple classification models
    let models = vec![
        "distilbert-base-uncased-finetuned-sst-2-english",
        "cardiffnlp/twitter-roberta-base-sentiment-latest",
        "nlptown/bert-base-multilingual-uncased-sentiment",
    ];

    println!("Creating ensemble with models:");
    for model in &models {
        println!("  - {}", model);
    }

    // Simulate ensemble predictions
    let test_sentence = "This movie is absolutely fantastic!";
    println!("\nTesting sentence: \"{}\"", test_sentence);

    // Simulate individual model predictions
    let predictions = vec![
        ("Model 1", "POSITIVE", 0.92),
        ("Model 2", "POSITIVE", 0.88),
        ("Model 3", "POSITIVE", 0.94),
    ];

    println!("\nIndividual predictions:");
    for (model, label, confidence) in &predictions {
        println!("  {}: {} ({:.1}%)", model, label, confidence * 100.0);
    }

    // Calculate ensemble result
    let avg_confidence: f32 =
        predictions.iter().map(|(_, _, c)| c).sum::<f32>() / predictions.len() as f32;
    println!("\nEnsemble result:");
    println!("  Label: POSITIVE");
    println!("  Confidence: {:.1}%", avg_confidence * 100.0);
    println!("  Voting: Unanimous (3/3)");

    Ok(())
}

/// Demonstrate conversational pipeline with memory
async fn conversational_example() -> Result<()> {
    println!("💬 Conversational Pipeline Example");
    println!("==================================");

    // Simulate a conversational pipeline with context
    let conversation_history = vec![
        ("user", "Hello! What's the weather like?"),
        ("assistant", "I'm sorry, I don't have access to current weather data. Is there something else I can help you with?"),
        ("user", "Can you tell me about machine learning?"),
        ("assistant", "Machine learning is a subset of artificial intelligence that enables computers to learn and improve from experience without being explicitly programmed."),
        ("user", "What about deep learning?"),
    ];

    println!("Conversation history:");
    for (role, message) in &conversation_history {
        println!("  {}: {}", role, message);
    }

    // Simulate context-aware response generation
    println!("\nGenerating context-aware response...");
    let response = "Deep learning is a subset of machine learning that uses neural networks with multiple layers to model and understand complex patterns in data. It's particularly effective for tasks like image recognition, natural language processing, and speech recognition.";

    println!("assistant: {}", response);

    // Show conversation features
    println!("\nConversational features:");
    println!("  ✓ Context awareness");
    println!("  ✓ Memory management");
    println!("  ✓ Coherent responses");
    println!("  ✓ Multi-turn dialogue");

    Ok(())
}

/// Demonstrate custom pipeline composition with custom logic
async fn custom_composition_example() -> Result<()> {
    println!("⚙️  Custom Pipeline Composition Example");
    println!("=======================================");

    // Custom composition example: Content Analysis Pipeline
    println!("Creating custom content analysis pipeline...");

    // Simulate a complex content analysis workflow
    let content = "Artificial intelligence is revolutionizing many industries. \
                   However, there are concerns about job displacement and ethical implications. \
                   Despite these challenges, AI offers tremendous opportunities for innovation \
                   and solving complex problems.";

    println!("Input content: \"{}\"", content);

    // Step 1: Text preprocessing
    println!("\n1. Text Preprocessing:");
    println!("   - Tokenization: ✓");
    println!("   - Sentence segmentation: ✓");
    println!("   - Language detection: English (99.2%)");

    // Step 2: Sentiment analysis
    println!("\n2. Sentiment Analysis:");
    println!("   - Overall sentiment: Neutral-Positive");
    println!("   - Positive aspects: 'opportunities', 'innovation'");
    println!("   - Concerns: 'job displacement', 'ethical implications'");

    // Step 3: Topic extraction
    println!("\n3. Topic Extraction:");
    println!("   - Primary topic: Artificial Intelligence (87%)");
    println!("   - Secondary topics: Technology Impact (23%), Ethics (19%)");

    // Step 4: Key phrase extraction
    println!("\n4. Key Phrase Extraction:");
    let key_phrases = vec![
        "artificial intelligence",
        "job displacement",
        "ethical implications",
        "innovation opportunities",
        "complex problems",
    ];
    for phrase in &key_phrases {
        println!("   - {}", phrase);
    }

    // Step 5: Content scoring
    println!("\n5. Content Scoring:");
    println!("   - Readability: 8.2/10");
    println!("   - Objectivity: 7.5/10");
    println!("   - Informativeness: 8.8/10");

    Ok(())
}

/// Utility functions for advanced composition

/// Custom pipeline that combines multiple operations
pub struct ContentAnalysisPipeline {
    sentiment_analyzer: Arc<Box<dyn Pipeline<Input = String, Output = PipelineOutput>>>,
    summarizer: Arc<Box<dyn Pipeline<Input = String, Output = PipelineOutput>>>,
}

impl ContentAnalysisPipeline {
    pub fn new() -> Result<Self> {
        let sentiment_analyzer = Arc::new(pipeline(
            "text-classification",
            Some("cardiffnlp/twitter-roberta-base-sentiment-latest"),
            None,
        )?);

        let summarizer = Arc::new(pipeline(
            "summarization",
            Some("facebook/bart-large-cnn"),
            None,
        )?);

        Ok(Self {
            sentiment_analyzer,
            summarizer,
        })
    }

    pub async fn analyze(&self, content: &str) -> Result<ContentAnalysisResult> {
        // Perform sentiment analysis
        let sentiment = self.sentiment_analyzer.__call__(content.to_string())?;

        // Generate summary
        let summary = self.summarizer.__call__(content.to_string())?;

        Ok(ContentAnalysisResult {
            original_text: content.to_string(),
            sentiment: format!("{:?}", sentiment), // Convert PipelineOutput to string
            summary: format!("{:?}", summary),     // Convert PipelineOutput to string
            word_count: content.split_whitespace().count(),
            character_count: content.len(),
        })
    }
}

/// Result structure for content analysis
#[derive(Debug)]
pub struct ContentAnalysisResult {
    pub original_text: String,
    pub sentiment: String, // In a real implementation, this would be a proper type
    pub summary: String,   // In a real implementation, this would be a proper type
    pub word_count: usize,
    pub character_count: usize,
}

/// Performance benchmarking for composition
#[allow(dead_code)]
pub async fn benchmark_composition() -> Result<()> {
    use std::time::Instant;

    println!("📊 Pipeline Composition Benchmarks");
    println!("==================================");

    let test_inputs: Vec<String> = (0..50)
        .map(|i| {
            format!(
                "This is test input number {} for benchmarking pipeline composition.",
                i
            )
        })
        .collect();

    // Single pipeline benchmark
    let classifier = pipeline(
        "text-classification",
        Some("distilbert-base-uncased-finetuned-sst-2-english"),
        None,
    )?;

    let start = Instant::now();
    for input in &test_inputs[..10] {
        let _ = classifier.__call__(input.clone())?;
    }
    let single_time = start.elapsed();

    println!("Single pipeline (10 inputs): {:?}", single_time);
    println!("Average per input: {:?}", single_time / 10);

    // Batch processing
    let start = Instant::now();
    let _ = classifier.batch(test_inputs[..10].to_vec())?;
    let batch_time = start.elapsed();

    println!("Batch processing (10 inputs): {:?}", batch_time);
    println!(
        "Speedup: {:.2}x",
        single_time.as_nanos() as f64 / batch_time.as_nanos() as f64
    );

    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;

    #[tokio::test]
    async fn test_content_analysis_result() {
        let result = ContentAnalysisResult {
            original_text: "test".to_string(),
            sentiment: "positive".to_string(),
            summary: "test summary".to_string(),
            word_count: 1,
            character_count: 4,
        };

        assert_eq!(result.word_count, 1);
        assert_eq!(result.character_count, 4);
    }
}