MLCA Week 8:

class: center, middle, inverse, title-slide

# MLCA Week 8:
## GBM Followup
### Mike Mahoney
### 2021-10-20

---

class: middle

# Project FAQ

---

# How much data do I need?

Enough -- but not too much.

</div>

<br />

Actually though: enough that your training data and test data both look like
the overall population; not so much that CV takes forever.

</div>

<br />

Fine: for this class, more than 100 rows, less than 20,000 rows. In the real 
world, tune your model on a subsample of your data then fit the final model on
the full data set (and prepare to run your computer for a few days).

</div>

---

# Do I need to use hand-coded bagged/boosted trees?

No.

</div>

<br />

If you're using random forests: just use ranger. If you're using GBM, just
use lightgbm (next week). If you're using SVM, just use kernlab (week 11).

</div>

<br />

We're coding through example implementations to try and explain concepts, but
there's no reason to ever use those again.

</div>

---
class: middle

After next week, you could be 100% done with the project.

</div>

<br />

Remember: if you want confirmation you checked all the boxes, I'm willing to grade each project <i>twice</i>.

</div>

<br />

To submit: email me (mike.mahoney.218@gmail.com) data & code with subject line "FOR796 Course Project".

</div>

<br />

Remember: your code needs to run on my computer immediately, without needing me to edit things. Use "here" and R Studio projects.

</div>

---

# What's left?

## • Week 9: Stochastic GBMs & Stacked Ensembles (long)

## • Week 10: K-Nearest Neighbors (normal-length)

## • Week 11: Support Vector Machines (normal-length, mathy)

## • Week 12 & 13: Project Work (optional)

---

# Presentations

Week 14 (2021-12-08).

</div>

<br />

Rubric: "Presentation included objectives, methods, results, and reflection."

</div>

<br />

Goal is to practice doing ML and talk about doing ML. Not meant to be high-stakes!

</div>

<br />