NeuroCraft Interface

NeuroCraft is Neurogebra's enhanced educational interface — like MathForge but with built-in learning tools, tutorials, and smarter error handling.

---

NeuroCraft vs MathForge

Feature	MathForge	NeuroCraft
Get expressions	✅	✅
Search	✅	✅ Smart with suggestions
Educational mode	❌	✅ Built-in explanations
Tutorials	❌	✅ Interactive tutorials
"Did you mean?"	❌	✅ Typo correction
Compare expressions	❌	✅ Side-by-side comparison
Quick access	❌	✅ quick_activation(), quick_loss()

---

Getting Started

from neurogebra import NeuroCraft

# Educational mode ON (default) — shows tips and explanations
craft = NeuroCraft(educational_mode=True)
# Output:
# 🎓 Welcome to Neurogebra!
#    Type craft.tutorial() to start learning
#    Type craft.search('activation') to explore

# Quiet mode — for experienced users
craft = NeuroCraft(educational_mode=False)

---

Getting Expressions

# Standard get
relu = craft.get("relu")

# With instant explanation
sigmoid = craft.get("sigmoid", explain=True)
# Automatically prints explanation when educational_mode=True

# With parameter override
leaky = craft.get("leaky_relu", params={"alpha": 0.2})

Smart Error Handling

# If you mistype:
try:
    craft.get("rellu")  # Typo!
except KeyError as e:
    print(e)
# Expression 'rellu' not found.
#    Did you mean: relu, leaky_relu, gelu?
#    Use craft.search('rellu') to find related expressions.

---

Searching with NeuroCraft

# Search with detailed output
results = craft.search("activation")
# Prints a formatted list with descriptions

# Search by category
results = craft.search("loss", category="loss")

# Search specific terms
results = craft.search("smooth")
results = craft.search("classification")

---

Quick Access Methods

# Quick activation — get + explain in one call
relu = craft.quick_activation("relu")

# Quick loss
mse = craft.quick_loss("mse")

# These automatically show explanations in educational mode

---

Comparing Expressions

# Visual comparison of activations
craft.compare(
    ["relu", "sigmoid", "tanh", "swish"],
    metric="behavior"
)

# Compare gradients
craft.compare(
    ["relu", "sigmoid", "tanh"],
    metric="gradient"
)

---

Interactive Tutorials

# Show tutorial menu
craft.tutorial()

# Start a specific tutorial
craft.tutorial("basics")
craft.tutorial("activations")
craft.tutorial("training")
craft.tutorial("first_model")

---

Listing All Expressions

# All expressions
all_exprs = craft.list_all()
print(f"Total: {len(all_exprs)}")

# By category
print("Activations:", craft.list_all(category="activation"))
print("Losses:", craft.list_all(category="loss"))
print("Regularizers:", craft.list_all(category="regularizer"))

---

Composing with NeuroCraft

# String composition
hybrid_loss = craft.compose("mse + 0.1*mae")

# Evaluate
result = hybrid_loss.eval(y_pred=3.0, y_true=5.0)
print(f"Loss: {result}")

---

Registering Custom Expressions

from neurogebra import Expression

# Create your own
my_activation = Expression(
    "xtanh",
    "x * tanh(x)",
    metadata={
        "category": "activation",
        "description": "x*tanh(x) — smooth activation"
    }
)

# Register it
craft.register("xtanh", my_activation)

# Now available everywhere
retrieved = craft.get("xtanh")
print(retrieved.eval(x=2.0))

---

When to Use NeuroCraft vs MathForge

Scenario	Use
Learning/exploring	NeuroCraft with educational_mode=True
Production code	MathForge (lighter, no prints)
Interactive notebooks	NeuroCraft
Scripts and pipelines	MathForge
Teaching others	NeuroCraft

---

Next: Datasets →