Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Gurneet1928/TryMLEasy/llms.txt

Use this file to discover all available pages before exploring further.

TryMLEasy builds a tf.keras.Sequential model dynamically from a layer table you fill in on the Neural Network Configurations page. You define the hidden layers; the app automatically prepends an Input layer sized to your feature set and appends a fixed Dense(1) output layer. Once the architecture is verified on the Neural Network Architecture page, training hyperparameters (epochs and batch size) are set on the Training Neural Networks page. This reference documents every option available across those three pages.

​
Model Architecture

The model is assembled as a tf.keras.Sequential stack with three parts:
  1. Input layer — added automatically; shape is set to the number of features selected on the Feature Selection page (xtrain.shape[1]).
  2. Hidden layers — one or more Dense layers defined by you in the layer editor table.
  3. Output layer — added automatically as Dense(1) (single neuron, no activation).
The full construction and compilation code used by TryMLEasy is: 
model = tf.keras.Sequential()
model.add(tf.keras.layers.Input(shape=(num_features,)))

# User-defined Dense layers, one per row in the layer table
model.add(tf.keras.layers.Dense(neurons, activation=activation))
# ... additional layers ...

model.add(tf.keras.layers.Dense(1))  # output layer, added automatically

model.compile(optimizer='adam', loss='mse', metrics=['mse'])
After compilation, TryMLEasy displays a full model.summary() on the Architecture Verification page so you can inspect parameter counts before training.
The output layer is always Dense(1) with no activation — TryMLEasy currently supports single-value regression targets only. Multi-output regression and classification neural networks are not supported in the current version.

​
Layer Configuration

Hidden layers are defined in a dynamic table on the Neural Network Configurations page. Each row in the table represents one Dense layer added to the model in order. The table has the following columns:
ColumnTypeConstraintsDefaultDescription
Layer NameTextRequiredA Happy LayerA display label for the layer. Has no effect on model behaviour — used for your own reference only.
NeuronsIntegerMin: 2, Max: 100, step: 1, required10Number of units in the Dense layer.
Activation FunctionSelectrelu, sigmoid, tanh, requiredreluThe activation function applied element-wise after the linear transformation.
Rows can be added or removed dynamically using the table controls. Layers are added to the model in the order they appear in the table (top to bottom = first to last hidden layer).
Index requirement: TryMLEasy requires you to manually set the row index when adding each new layer. Rows without an explicit index value will not be accepted when you click “Verify Architecture”. This is a known UX limitation — the team is actively working to improve this experience.

​
Activation Functions

Rectified Linear Unit
f(x) = max(0, x)
ReLU outputs the input directly if it is positive, and zero otherwise. It is the default and recommended choice for hidden layers because it:
  • Introduces non-linearity without saturating for positive inputs.
  • Helps avoid the vanishing gradient problem that affects sigmoid and tanh in deep networks.
  • Is computationally cheap to evaluate and differentiate.
Recommended for: Most hidden layers in regression networks.

​
Training Hyperparameters

Training hyperparameters are set on the Training Neural Networks page:
ParameterInput TypeMinimumDescription
EpochsInteger1Number of complete passes over the full training dataset. More epochs allow the model to learn longer but increase risk of overfitting.
Batch SizeInteger1Number of training samples processed before performing a single weight update. Smaller batches produce noisier but more frequent updates; larger batches are smoother but require more memory.
OptimizerFixed—Adam — adaptive learning rate optimiser. Not configurable in the current version.
Loss FunctionFixed—Mean Squared Error (MSE) — average of squared differences between predictions and true values. Not configurable in the current version.
Validation dataAutomatic—The xtest / ytest split produced by the Feature Selection page is used as the validation set at every epoch.
The “Train Neural Network” button is disabled until both Epochs and Batch Size are set to a value greater than zero.

​
Training Output

While training runs, TryMLEasy streams per-epoch metrics to the page in real time using a custom Keras callback, then renders a final summary and plot.

​
Per-Epoch Log

Each epoch appends one line to the page: 
Epoch ---- Loss ---- Val_Loss
0 ----- 0.42 ----- 0.51
1 ----- 0.38 ----- 0.47
...
ColumnDescription
EpochZero-indexed epoch number.
LossMSE on the training split for that epoch.
Val_LossMSE on the test (validation) split for that epoch.

​
Final Test Evaluation

After all epochs complete, the model is evaluated on the test set one final time: 
Final Loss on Test Data: <value>

​
Training vs. Validation Loss Plot

A matplotlib line chart is displayed showing both train_loss and val_loss over all epochs. Use this plot to:
  • Confirm that training loss is decreasing — if it is not, try more epochs or a larger network.
  • Detect overfitting: a growing gap where val_loss rises while train_loss continues to fall.
  • Detect underfitting: both losses remain high — try adding more neurons or layers.

Build docs developers (and LLMs) love

Get started for freeTalk to us