Every block you need to build ML pipelines

Load structured data from CSV files for analysis and model training.

Import data from Excel spreadsheets (.xlsx, .xls) including multi-sheet files.

Load columnar-format Parquet files — ideal for large datasets.

Connect to relational databases (PostgreSQL, MySQL, SQLite) and run custom queries.

Fetch data from REST APIs with support for authentication and pagination.

Handle missing values through deletion or statistical imputation.

Advanced missing value imputation using KNN or iterative model-based strategies.

Normalise or standardise numeric features to comparable scales.

Convert categorical variables to numeric representations for ML models.

Synthetic Minority Over-sampling Technique for imbalanced classification.

Detect and remove or cap statistical outliers from numeric features.

Divide dataset into training and test sets with optional stratification.

Principal Component Analysis — reduce dimensionality while preserving variance.

Replace categorical values with smoothed mean of the target variable.

Generate interaction terms and polynomial combinations of numeric features.

Ensemble of decision trees — robust, accurate, and scale-invariant.

Optimised gradient boosting — frequently wins structured-data competitions.

Gradient boosting optimised for speed and large datasets.

Linear classifier with probabilistic output — fast and interpretable.

Support Vector Machine — effective in high-dimensional spaces.

Interpretable tree-based model — transparent decision rules.

Multi-layer perceptron for complex non-linear patterns in tabular data.

Long Short-Term Memory — recurrent network for sequential tabular data.

Self-attention based model — state of the art for tabular and sequence tasks.

Classification accuracy, precision, recall, and F1 — all in one block.

Area under the ROC curve — threshold-independent classifier evaluation.

Harmonic mean of precision and recall — best for imbalanced classification.

Visualise all correct and incorrect predictions across every class.

Visualize pairwise correlations between numeric features as a color-coded heatmap matrix.

Display pairwise correlations in a sortable table format with numeric precision.

Game-theory feature importance — explains how each feature drives predictions.

K-fold cross-validation for robust, unbiased performance estimates.

Load image datasets from a folder organised by class subdirectories.

Load object detection datasets in COCO JSON annotation format.

Resize images to a fixed resolution required by vision models.

Normalise pixel values to ImageNet mean/std for pretrained model compatibility.

Data augmentation via horizontal and vertical random flipping.

Efficient and accurate image classifier — best accuracy/parameter ratio.

Classic 50-layer residual network — reliable baseline for image tasks.

Real-time object detection — state of the art speed and accuracy.

Mean Average Precision — the standard metric for object detection.

Load text data from CSV files for NLP pipelines.

Extract text from PDF files for document analysis pipelines.

Remove noise from text — HTML, URLs, special characters, stopwords.

Convert text to numeric TF-IDF feature vectors for classical ML models.

Dense sentence embeddings using sentence-transformers models.

Bidirectional transformer — state of the art for text classification and NER.

Lightweight BERT distillation — 40% smaller, 60% faster, 97% of performance.

Connect to GPT-4, Claude, or Llama via API for zero-shot NLP tasks.

Load time series data from CSV with automatic datetime parsing.

Stream historical stock, crypto, or forex prices from market data APIs.

Change time series frequency — upsample or downsample with aggregation.

Remove trends and make time series stationary via differencing.

Create lagged versions of the target to use as input features.

Rolling mean, std, min, max over a sliding time window.

Facebook Prophet — robust forecasting with trend, seasonality, and holidays.

AutoRegressive Integrated Moving Average — classic statistical forecasting.

XGBoost trained on lag and rolling features for time series regression.

Mean Absolute Error — average magnitude of forecast errors.

Root Mean Squared Error — penalises large forecast errors more heavily.

Load vector geospatial data from Shapefile or GeoJSON.

Load GeoJSON feature collections directly from file or URL.

Transform geometries between coordinate reference systems (CRS).

Sample spatial data onto a regular grid at a specified resolution.

Compute vegetation indices (NDVI, EVI) from multispectral satellite bands.

Compute distances from points to reference geometries (roads, cities, POIs).

Random Forest with spatial cross-validation to prevent leakage.

Gaussian process spatial interpolation — optimal linear unbiased estimator.

Evaluate predictions with spatially-aware metrics (RMSE, coverage error).