heri-eval(1)

💖 heri-eval — the manual page, but make it make sense 💖

okay bestie, welcome to heri-eval(1) ✨ this is the lil command that trains a classification algorithm on your dataset and then judges it (affectionately). Let's get into it 💅

---

✨ NAME ✨

heri-eval — evaluate classification algorithm 🎀

---

🌸 SYNOPSIS 🌸

heri-eval [OPTIONS] dataset [-- SVM_TRAIN_OPTIONS]

---

🦋 DESCRIPTION 🦋

So here's the vibe: heri-eval runs the training algorithm on your dataset, and then it evaluates the whole thing using a testing set — which you hand it with the -e option, no cap.

If you used option -n, then cross-validation is what's doing the evaluating, and the training and testing on different folds get run in parallel, so it's out here actually using the CPUs you have available. main character energy fr fr 🔥

And if -r is in the mix, the dataset gets split into training and testing datasets randomly, using the ratio you specify, and then holdout is run. periodt. 💅

---

👑 OPTIONS 👑

• -h, --help — shows you the help information. the tutorial level 🫶
• -f — turns on output of per-fold statistics. peep -Mf for the deets 👀
• -n N — enables T*N-fold cross-validation mode and sets the number of folds to N. slay.
• -r ratio — splits the dataset into training and testing parts with the specified ratio of their sizes (in percents).
• -t T — enables T*N-fold cross-validation mode and sets the number of runs to T, which is 1 by default. one run, one vibe, unless you say otherwise ✨
• -e testing_dataset — enables hold-out mode and sets the testing dataset.
• -T threshold — sets the minimum threshold for making a classification decision. If this flag is applied, micro-average precision, recall, and F1 are calculated instead of accuracy. different metrics, same drama 💅
• -o filename — saves predictions from testing sets to the specified file.

Format: outcome_class prediction_class [score]

• -O filename — saves incorrectly classified objects to the specified file. (the flops, we keep receipts 🍓)

Format: #object_number: outcome_class prediction_class [score])

• -m filename — saves the confusion matrix to the specified file. (yes it's literally called that, iconic name ngl)

Format: frequency : outcome_class prediction_class

• -p opts — passes the specified opts to heri-stat(1).
• -s opts — passes the specified opts to heri-split(1).
• -M chars — sets the output mode, where chars are: t — output total statistics, f — output per-fold statistics, c — output cross-fold statistics. The default is "-M tc". 🎀
• -S seed — passes the specified seed to heri-split(1). reproducibility queen 🌈
• -K — keeps the temporary directory after exiting. no cleanup, we're sentimental like that 🙈
• -D — turns on the debugging mode, and it implies -K. (debug brain on, temp dir stays 💀)

---

🌟 EXAMPLES 🌟

 heri-eval -e testing_set.libsvm training_set.libsvm -- -s 0 -t 0

 export SVM_TRAIN_CMD='liblinear-train'
 export SVM_PREDICT_CMD='liblinear-predict'
 heri-eval -p '-mr' -n 5 training_set.libsvm -- -s 4 -q
 heri-eval -p '-mr' -n 5 training_set.libsvm -- -s 4 -q

 export SVM_TRAIN_CMD='scikit_rf-train --estimators=400'
 export SVM_PREDICT_CMD='scikit_rf-predict'
 heri-eval -p '-c' -Mt -t 50 -r 70 dataset.libsvm

---

🧚 ENVIRONMENT 🧚

• SVM_TRAIN_CMD — the training utility, e.g., liblinear-train (the default is svm-train).
• SVM_PREDICT_CMD — the predicting utility, e.g., liblinear-predict (the default is svm-predict).
• SVM_HERI_STAT_CMD — the utility for calculating statistics (the default is heri-stat(1)).
• SVM_HERI_STAT_ADDONS_CMD — the utility for calculating additional statistics (the default is heri-stat-addons(1)).
• SVM_HERI_SPLIT_CMD — the utility for splitting the dataset (the default is heri-split(1)).
• TMPDIR — the temporary directory (the default is /tmp).

---

🏠 HOME 🏠

<http://github.com/cheusov/herisvm>

go say hi to the repo bestie 🫶

---

💜 SEE ALSO 💜

heri-split(1) heri-stat(1)

---

and that's the whole page babes 💖 this was the heri-eval(1) manual, dated 2021-01-25 — mwah, thanks for reading, you ate that 🎀✨👑