Command line¶

Every command has --help. Global options go before the command: amesml -v train ... shows debug logging, amesml -q train ... only warnings.

Command	Does
`amesml data download`	Download `data/train.csv` and verify its checksum
`amesml train`	Train a model, print held-out metrics, save it
`amesml evaluate`	Score a saved model on labelled data
`amesml predict`	Predict for every row of a CSV
`amesml plots`	Regenerate the figures from the data
`amesml serve`	Run the REST API

Errors are reported on one line with exit code 1; usage errors exit with 2.

train¶

amesml train                                   # the notebook's linear regression
amesml train --model gbm                       # gradient boosting
amesml train --task classification             # the notebook's logistic regression
amesml train --config examples/config.example.yaml --name my-model

The model is saved to models/<task>-<model>/ (change with --out and --name) as three files:

model.joblib - the fitted scikit-learn pipeline
metadata.json - configuration, metrics, feature defaults, library versions, a fingerprint of the training data and a checksum of model.joblib
metrics.json - the held-out metrics alone

--task and --model override the configuration file.

evaluate¶

amesml evaluate --model models/regression-gbm --data data/train.csv --split test

--split test recreates the exact rows the model was held out on, so it reproduces the metrics printed at training time. --split train scores the training rows, and --split all (the default) every row; amesml warns when those include rows the model was fitted on.

predict¶

amesml predict --model models/regression-gbm --input houses.csv > predictions.csv
amesml predict --model models/classification-gbm --input houses.csv --out classes.csv

The input uses the Ames column names; NA and empty cells mean missing. Columns the model does not use (such as SalePrice) are ignored, columns it needs but the file lacks are filled with typical training values, and an Id column is copied to the output. Regression writes predicted_price; classification writes predicted_label (1 = at or above the threshold) and probability_above_threshold.

plots¶

pip install "amesml[plots] @ git+https://github.com/Ahsan-muhd-444/house-price-prediction@v0.1.1"
amesml plots --out images/

Writes the sale-price distribution, missing values per column, residuals of the linear model, the classifier's confusion matrix and the gradient-boosting model's permutation importance, as images/amesml_*.png.

serve¶

amesml serve --model models/regression-gbm --model models/classification-gbm

Serves at most one model per task. Without --model, the comma-separated paths in AMESML_MODEL_PATH are used. --host and --port also read AMESML_HOST and AMESML_PORT. See REST API.

data download¶

amesml data download                      # to data/train.csv
amesml data download --out elsewhere/     # to elsewhere/train.csv
amesml data download --url URL --sha256 HASH

The default source is pinned to a commit of this repository and checked against a SHA-256 before anything is written.