Home / Tools
ML Tools
Free browser-based machine learning tools. Train neural networks, calculate metrics, find learning rates, estimate model complexity, and split datasets. Everything runs locally.
Neural Network Playground
Choose datasets, configure hidden layers, pick activations, and watch decision boundaries form in real time.
Confusion Matrix Calculator
Precision, recall, F1, accuracy, specificity, and MCC with an SVG heatmap, ROC point plotting, and multi-class support up to 5 classes.
Learning Rate Finder
Visualize loss vs learning rate on a log scale, auto-detect the optimal LR at steepest descent, compare warm restarts vs one-cycle scheduling.
Model Complexity Calculator
Build neural networks layer by layer to calculate parameter count, FLOPS, and memory footprint. Compare ResNet, VGG, BERT, GPT architectures.
Dataset Splitter Calculator
Configure train/validation/test ratios, preview stratified splits, check class balance, and generate Python code for sklearn train_test_split.
Attention Mechanism Visualizer
Enter a sentence to see self-attention weights as a heatmap. Toggle self vs cross-attention. Browser-only.
Loss Function Explorer
Compare MSE, MAE, Huber, Cross-Entropy, Hinge, Focal, Log-Cosh, and Quantile loss on interactive plots. Slider controls for parameters.
Activation Function Visualizer
Interactive chart for 18 activation functions with derivatives, formulas, PyTorch code, pros/cons and comparison mode.
Batch Normalization Calculator
Compute mean, variance, normalize, scale and shift. Visualize distributions with a canvas histogram.
Bias-Variance Tradeoff Calculator
Adjust polynomial degree, noise, and sample size to see training error, test error, bias², variance, and total error.
Confusion Matrix Analyzer
Calculate precision, recall, F1-score, MCC, Cohen's Kappa from confusion matrices. Binary and multi-class classification metrics.
Dropout Rate Optimizer
Pick the right dropout rate for dense, convolutional, and recurrent layers. Visualize dropped neurons and compute effective capacity.
Feature Importance Calculator
Calculate permutation feature importance. Input features and weights, see ranked importance scores and interactive charts.
Gradient Clipping Calculator
Clip by value, norm, or global norm. Visual comparison, recommended thresholds, PyTorch/TF code snippets.
Hyperparameter Search Calculator
Compare grid vs random search vs Bayesian optimization. Calculate search space coverage and find the optimal strategy.
Learning Rate Scheduler Visualizer
Visualize cosine annealing, step decay, and warmup schedules. Compare strategies and generate PyTorch code.
Model Complexity Estimator
Estimate model FLOPs, memory usage, and inference time. Compare neural network architectures instantly.
Model Ensemble Calculator
Combine up to 10 models via averaging, voting, or stacking. Compute Fleiss' kappa, agreement, and a diversity-based accuracy estimate.
Optimizer Trajectory Visualizer
Watch SGD, Momentum, Adam and RMSprop descend a 2D loss surface. Step through the update math and compare convergence paths.
Overfitting Detector
Input train/val loss curves, detect the overfitting point, and get regularization suggestions.
ROC & PR Curve Threshold Calculator
Paste prediction scores and labels to plot ROC and precision-recall curves, compute AUC, and find the optimal threshold by F1 or Youden's J.
ROC Curve Generator
Plot ROC curves and calculate AUC from prediction data. Compare model performance with sensitivity/specificity analysis.
Train Test Split Calculator
Calculate optimal train-test splits. Get stratification impact, k-fold recommendations, and sample size analysis.
Transformer FLOPs & Memory Calculator
Compute training/inference FLOPs, parameter count, activation memory and KV-cache size from layers, d_model, heads, seq length and batch.
Calculators
KV Cache Size Calculator
Per-token KV cache size for 19 open models, read from each model's config.json, with a custom-model path and the GQA over-report correction.
LLM Memory Calculator
Size LLM weights, KV cache, gradients and optimizer states from presets that carry exact fetched parameter counts.