tensor-talezz-rv 0.1.0

Making AI Approachable, Interpretable, and Resource-Friendly.

TensorTalezz-RV

Accessible, educational, and practical AI/ML tools for everyone.

---

🌟 Overview

TensorTalezz-RV is a proprietary AI/ML library designed to make machine learning approachable, educational, and practical. Structured into five core modules, it combines powerful, easy-to-use ML tools with interactive educational features to bridge the gap between abstract concepts and real-world implementation.

---

📦 Modules & Features

1. 🔍 Explainability

Interpret and trust your models with intuitive visualizations.

• plot_feature_influence: Visualize how each feature contributes to predictions.
• plot_decision_boundary: Show classifier boundaries in two dimensions.
• explain_prediction: Provide a human-readable breakdown of a single prediction.
• compare_models: Contrast multiple models side by side.
• plot_training_curve: Track loss and accuracy trends during training.

2. ⚡ Lightweight ML

Simplicity and speed for rapid prototyping.

• train: One-line model training with minimal boilerplate.
• predict: Generate predictions effortlessly.
• evaluate: Print clear, friendly, and comprehensive performance metrics.
• save_model & load_model: Persist and reuse trained models easily.

3. 🧹 Preprocessing

Automate the messy work of data preparation.

• auto_clean: Handle missing values, scaling, and encoding automatically.
• detect_outliers: Flag anomalies in datasets.
• normalize_text: Clean text data for NLP tasks.
• split_dataset: Perform train/test splits in one call.
• feature_summary: Generate quick statistics for each column.

4. 🎓 Education

Interactive tools to teach and learn AI concepts tangibly.

• explain: Analogy-driven explanations of complex AI/ML concepts.
• visualize_algorithm: Step-by-step algorithm demonstrations.
• concept_demo: Toy-dataset demonstrations of ideas like over/underfitting.
• quiz: Generate short, interactive quizzes for learners.
• compare_algorithms: Run multiple algorithms and explain their differences.

5. 📱 Low-Resource ML

Built for constrained environments with limited data or compute.

• tiny_classifier: Train highly constrained, lightweight models.
• quantize_model: Reduce model serialization size for deployment.
• evaluate_small_data: Use robust metrics tailored for tiny datasets.
• stream_train: Support incremental, out-of-core training for streaming data.
• compress_dataset: Reduce dataset size via intelligent downsampling or dimensionality reduction.

---

🛠 Requirements

• Python 3.8+
• Dependencies: numpy, pandas, scikit-learn, matplotlib, seaborn

---

🚀 Installation

# Clone the repository
git clone https://github.com/example/TensorTalezz-RV.git
cd TensorTalezz-RV

# Install the package
pip install .

---

🎯 Quick Start

Check out the examples/quickstart.py file to see a complete workflow combining all modules.

from tensor_talezz_rv.lightweight import train, evaluate
from tensor_talezz_rv.education import explain

# Learn about an algorithm
explain("random_forest")

# Train it in one line!
model = train(X_train, y_train, algorithm="random_forest")
metrics = evaluate(model, X_test, y_test)

---

📚 Usage Examples

Preprocessing

from tensor_talezz_rv.preprocessing import auto_clean

cleaned_df, report = auto_clean(raw_dataset)

Explainability

from tensor_talezz_rv.explainability import explain_prediction

feature_names = [f"Feature {i}" for i in range(X_test.shape[1])]
explanation = explain_prediction(model, X_test[0], feature_names)

Low-Resource ML

from tensor_talezz_rv.lowresource import tiny_classifier

tiny_model = tiny_classifier(X_train, y_train)

---

tensor_talezz_rv in Action

====================================================================== 🚀 WELCOME TO TENSORTALEZZ-RV LIVE DEMO

---

🧹 MODULE 1: PREPROCESSING

1a. Before Cleaning (Feature Summary):

📋 Feature Summary (250 rows, 6 columns)

column   dtype  missing  missing_pct  unique    mean      std       min         max    median 
    top_values
   Age float64        1          0.4     249 -0.0396   1.3134 -3.422492    3.370100 -0.229298 
           NaN
Income float64        1          0.4     249  0.0700   1.0655 -2.364563    2.367717  0.057826 
           NaN
 Score float64        0          0.0     250 39.9405 632.3972 -3.959492 9999.000000 -0.217356 
           NaN

Engagement float64 0 0.0 250 -0.0696 1.3463 -2.920976 3.498072 -0.041329 NaN Category object 86 34.4 2 NaN NaN NaN NaN NaN {'B': 83, 'A': 81} Target int64 0 0.0 2 0.5000 0.5010 0.000000 1.000000 0.500000 NaN

1b. Outlier Detection: 🔍 Detected 1 outlier(s) across 5 column(s). • Score: 1 outliers

1c. Auto Clean Action: 🩹 Filled missing: numeric=['Age', 'Income'], categorical=['Category'] 📏 Scaled numeric columns: ['Age', 'Income', 'Score', 'Engagement', 'Target'] 🏷️ Encoded categorical columns: ['Category'] ✅ Cleaning complete. Shape: (250, 6) ✂️ Split: 200 train / 50 test (test_size=0.2)

---

⚡ MODULE 2: LIGHTWEIGHT ML

2a. One-Line Training: ✅ Trained Random Forest (classification) on 200 samples. ✅ Trained Logistic Regression (classification) on 200 samples.

2b. Easy Evaluation:

📊 ── Evaluation Results ────────────────── accuracy : 0.9600 precision : 0.9628 recall : 0.9600 f1_score : 0.9598

          precision    recall  f1-score   support

    -1.0       1.00      0.91      0.95        23
     1.0       0.93      1.00      0.96        27

accuracy                           0.96        50

macro avg 0.97 0.96 0.96 50 weighted avg 0.96 0.96 0.96 50

──────────────────────────────────────────

2c. Saving Model: 💾 Model saved to 'demo_random_forest.joblib' (348.9 KB)

---

🔍 MODULE 3: EXPLAINABILITY

3a. Explaining a Single Prediction: Prediction: 1.0 --- Feature Contributions ---

• Age: 0.1445

• Income: -0.0931

• Engagement: 0.0195
• Category: 0.0087

• Score: -0.0076

3b. Comparing Models On Test Set:

Model Accuracy Precision Recall F1 Score

Random Forest 0.9600 0.9628 0.9600 0.9598 Logistic Regression 0.9200 0.9232 0.9200 0.9201

3c. Plotting Feature Influence (Close window to proceed):

---

📱 MODULE 4: LOW-RESOURCE ML

4a. Compressing Dataset (e.g. for Edge Devices): 🗜️ Original shape: (200, 5) 📉 Downsampled rows to 50. 📉 PCA reduced features to 2 components. ✅ Compressed shape: (50, 2)

4b. Small Data Evaluation (Leave-One-Out CV): 🚀 Trained 'fast_tree' tiny classifier on 15 samples. 🔬 Evaluating on small dataset (15 samples) using Leave-One-Out CV...

📊 ── Small Data Evaluation ────── LOO Accuracy : 0.8000 Balanced Accuracy : 0.7946 (Handles class imbalance) MCC : 0.6001 (Robust score -1 to +1) ──────────────────────────────────

---

🎓 MODULE 5: EDUCATION

5a. Explaining Concepts Analogy-Style:

============================================================ 📖 Bias–Variance Tradeoff

🎯 Analogy: Think of throwing darts. High bias means you consistently hit the wrong spot (systematic error). High variance means your throws are scattered all over the board (inconsistent). The sweet spot is low bias AND low variance — consistently hitting the bullseye.

📐 Formal Definition: Bias is the error from wrong assumptions; variance is sensitivity to small fluctuations in the training set. Total error = Bias² + Variance + Irreducible error. Reducing one often increases the other.

💡 Tips: • Complex models → low bias, high variance. • Simple models → high bias, low variance. • Use ensemble methods to balance both.

5b. Visualizing Algorithms Step-by-Step (Close window to exit): ⭐ The gold star is the query point. Notice how k affects the prediction and boundary smoothness.

====================================================================== 🎉 DEMO COMPLETE!

📄 License

This project is licensed under the TensorTalezz-RV Proprietary License.
See the LICENSE file for details.

---

💼 Commercial Use

TensorTalezz-RV is proprietary software.
For commercial licensing, enterprise support, integration into paid services, or collaborations, please refer to the COMMERCIAL.md file for full details and constraints.

To request a commercial license, please contact:
📧 [e-mail]

---

🤝 Contributing

Bug reports, feature requests, and suggestions are welcome.
Code modifications require explicit approval due to the proprietary license.