heri-split(1)

🦋 heri-split 🦋

okay so welcome to heri-split(1), the little tool that takes your dataset and goes snip snip into training and testing sets ✂️✨ if you're doing N-fold cross-validation, this is literally your girl. she's part of the herisvm family and she is, ngl, kind of iconic.

---

🌸 NAME 🌸

heri-split — splits the dataset into training and testing sets 🎀

---

✨ SYNOPSIS ✨

heri-split [OPTIONS] dataset1 [dataset2...]

like... you give her options, you give her one dataset, or a bunch of datasets, and she does her thing 💅

---

💜 DESCRIPTION 💜

so here's the vibe: heri-split splits your dataset into several training and testing sets, exactly how you need it for N-fold cross-validation 🧚 the dataset has one object per line, in svmlight format, no cap.

by default she uses stratified sampling — which means every fold gets the same number of objects for each label. balanced. fair. main character behaviour fr fr 👑

now, the outputs, and this part matters bestie, so listen up:

• if you pass the -c option, she creates testN.txt and trainN.txt files (also in svmlight format, obviously 💖), where N is the number of the fold.
• if you pass the -R option, she creates test.txt and train.txt files instead, for the same purposes.
• also she creates a testing_fold.txt file 🍓 — for each object (one per line) it tells you that object's testing fold number, if the option -c is applied.
• and if the option -R is applied, that same testing_fold.txt file contains either 1 for testing set or 0 for training set.

💀 real talk: -c is mandatory. -d is mandatory. she will not run without them. she's not being difficult, she's just like that.

---

🎀 OPTIONS 🎀

• -h, --help — display help information 🫶 she'll tell you what's up.
• -c, --folds count — set the number of folds. this one is a mandatory option, bestie. no -c, no folds, no service.
• -d, --output-dir dir — set the output directory. also mandatory. she needs to know where to drop her files 💅
• -r, --random — use random sampling instead of the stratified one 🎲 going feral, going unhinged, love that for you.
• -R, --ratio — split the input dataset into a training one and a testing one in the specified ratio (in percents) 🌈
• -s, --seed seed — set the seed value for the pseudorandom generator 🌱 so your randomness is reproducible. consistent queen.

---

🏠 HOME 🏠

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

---

👀 SEE ALSO 👀

heri-eval(1) heri-stat(1)

---

and that's the whole page, bestie 💖 this was heri-split(1), from the 2021-01-25 era, and she still eats. now go split something 🦋✨ periodt.