heri-stat(1)

🌸 heri-stat(1) — bestie edition 🌸

okay so heri-stat is the little command that does your math for you when you're evaluating a model, and it's genuinely so good at it. 💖 It crunches your outcomes against your predictions and hands you precision, recall, F1 and a whole lil squad of other numbers. no cap.

---

✨ NAME ✨

heri-stat — calculates precision, recall, F1 and some other things 💅

---

🎀 SYNOPSIS 🎀

like, here's every single way you can call her, keep them all:

heri-stat [-R] [-mrca] [-u label] [-t threshold]
   outcomes_file predictions_file

heri-stat -1 [-R] [-mrca] [-u label] [-t threshold]
   [files...]

heri-stat -g mode [-R] [-xy]  outcomes_file predictions_file

heri-stat -1 -g mode [-R]  [files...]

heri-stat [-h]

---

🦋 DESCRIPTION 🦋

so the first and second flavors of calling heri-stat are the classification girlies. you feed her a classification dataset plus predictions, and she pops out precision, recall and F1. periodt.

unless you used the -1 option, she reads the correct classes out of outcomes_file — one class per line, okay, literally one per line — and the predicted classes out of predictions_file, also one class per line. 🙈

and here's the fun part: predictions_file is allowed to have two tokens per line. first token = the predicted class, second token = a score (like, a probability, y'know). she handles it. she's versatile.

---

now the third and fourth flavors are the regression era. you give her regression outcomes and predictions — one value per line — and she calculates mean absolute error (MAE), mean squared error (MSE) and/or mean absolute error (MAE). yes bestie, that's exactly what it says, I'm not touching it, I just report the facts. 😭

and if -1 was applied, she expects two or three tokens per line on input:

• correct value (or class, if you're doing classification) 🍓
• predicted value (or class) ⭐
• and an optional score 🧚

---

💅 OPTIONS 💅

• -h, --help — displays the help info. the "help me bestie" button. 🫶
• -R, --raw — raw tab-separated output. no glow-up, just vibes.
• -m, --micro-avg — disables the micro averaged P/R/F1 output.
• -r, --macro-avg — disables the macro averaged P/R/F1 output.
• -c, --per-class — disables the per-class statistics output.
• -a, --accuracy — disables the accuracy output.
• -1, --single — 2 or 3 tokens per line are expected on input. that's the whole thing. 💀
• -u, --unclassified label — sets the label for the "unclassified" object. if you specify it, micro-averaged P/R/F1 is calculated instead of accuracy. also, the statistics for this class is not calculated. she's off the roster. 🙈
• -t, --treshold value — treat classes with a score less than the specified value as unclassified. (yes, the option is spelled --treshold, that's canon, don't @ me, I copy exactly.) 🎀
• -g, --regression mode — regression outcomes and predictions are expected on input. if mode contains the symbol a, MAE is output; s → MSE is output; and r → RMSE is output. mix and match, she's flexible like that. 🔥

---

🌈 HOME 🌈

<http://github.com/cheusov/herisvm> — go say hi to the house. 👑

---

👑 SEE ALSO 👑

• heri-eval(1)
• heri-split(1)

the rest of the friend group, fr fr. 💜

---

and that's the whole page bestie, this version is from 2021-01-25 and it's the heri-stat(1) manual — go forth and compute, you're gonna slay. 💖✨🎀🦋🌸💅🫶