Machine Learning with PyTorch

Dataset (CSV/XLS/XLSX)

Training Algorithm

What it is: Neural Net (dense network): learns nonlinear feature interactions using hidden layers with backpropagation.

Why it's unique: Flexible general-purpose model that captures complex interactions a linear baseline can miss, and works well as your default deep tabular learner.

Distillation Note: Supported.

Train PyTorch Model

Epochs: 60

Batch sizes: 64

Learning rates: 0.001

Test sizes: 0.2

Hidden dims: 128

Hidden layers: 2

Dropouts: 0.1

Dataset: none

Planned runs: 1

Optional Settings

Bayesian Optimization

What is it: A method for optimizing expensive black-box functions by using a probabilistic model to choose promising parameter settings.

How it works: Uses completed runs to suggest the next promising hyperparameter combination. Requires at least 5 completed runs for the specific algorithm.

Toggle ON to apply sweep values. Use Reload for a fresh random sweep.

Training Runs

No runs yet. Train once to populate the results table.

Preprocessing Notes

Categorical Encoding: Text columns with ≤ 20 unique values are automatically One-Hot Encoded.

High Cardinality & IDs: Text columns with > 20 unique values or ID-like names are dropped to prevent feature explosion.

Date Parsing: Dates and timestamps are extracted into Year, Month, and Day numeric features.

Missing Values: Missing numeric values are imputed using the column median to maintain robustness against outliers.

Feature Scaling: All features are standardized to zero mean and unit variance (StandardScaler) before analysis. This prevents large-range features from dominating algorithms like PCA.

Dataset Table Preview

Showing 0 rows for no selection.