mathexec 0.1.0

Python SDK for MathExec — call published model endpoints

mathexec

Python SDK for MathExec — call your published model endpoints from code.

pip install mathexec

Quick start

from mathexec import Model

model = Model.load("your-model-id")
result = model.predict({"age": 25, "income": 50000})
print(result["predictions"])     # [1]
print(result["probabilities"])   # [0.92]

Get an API key

1. Sign in at mathexec.com
2. Open Settings → API keys
3. Create a key — it starts with mx_live_…. Copy it now; the full value is only shown once.
4. Pass it as api_key=, or set the MATHEXEC_API_KEY env var and read it yourself.

Public models don't need a key. Private models, rate-limited usage, and list_models() do.

Usage

Predict on one sample

from mathexec import Model

model = Model.load("project/experiment-3", api_key="mx_live_…")

# Named features
model.predict({"age": 25, "income": 50000})

# Positional features
model.predict([25, 50000])

Predict in batch

results = model.predict_batch([
    {"age": 25, "income": 50000},
    {"age": 45, "income": 80000},
])
results["predictions"]    # [1, 0]
results["probabilities"]  # [0.92, 0.31]

sklearn-style convenience

model.predict_label({"age": 25, "income": 50000})   # 'yes'
model.predict_proba({"age": 25, "income": 50000})   # 0.92

Model metadata

model.info["name"]            # "Churn classifier"
model.info["metrics"]         # {"accuracy": 0.92, "roc_auc": 0.94, ...}
model.task_type               # 'classification'
model.class_labels            # ['no', 'yes']
model.positive_class          # 'yes'

List your models

from mathexec import list_models

models = list_models(api_key="mx_live_…")
for m in models:
    print(m["model_id"], m["name"])

Self-hosted server

Model.load("id", base_url="http://localhost:8001", api_key="mx_live_…")

Errors

from mathexec import Model, MathExecError

try:
    model = Model.load("does-not-exist")
except MathExecError as e:
    print(e)   # "Model 'does-not-exist' not found."

MathExecError is raised for:

• 404 — model not found / not accessible
• 429 — rate limited
• 400 — bad input (e.g. wrong feature names or types)

For 5xx the underlying requests.HTTPError propagates.

License

MIT