Notes

from stone age befor llms

Python

VENV

1. make venv

python3 -mvenv .venv

2. to activate

source ./.venv/bin/activate

# install packages pip / pip3
pip install jupyter matplotlib numpy pandas scipy scikit-learn

# upgrade existing package
pip install --upgrade <package_name>

# make requirements.txt file
pip freeze > requirements.txt

# To exit venv
deactivate

# How to use the requirements.txt file
pip install -r requirements.txt

• To specify the interpreter you can use the #! on the first line of a script.py

#!/Users/mikedeufel/code/python/HuggingFace/venv/bin/python

• Note you must run this as an executable in order for the #! to work

chmod +x filename.py
ls -la #to verify
./filename.py

SQLite

import sqlite3
import pprint
''' Inspect '''
print(db.execute("select sqlite_version()").fetchall()) # sqlite version
db = sqlite3.connect("your_datebase.db")
db.execute("select name from sqlite_master where type='table'").fetchall()  # tables
for row in db.execute("SELECT * FROM any_table LIMIT 10"): print(row)       # rows (tuples)
db.row_factory = sqlite3.Row              # will convert rows to dict (easier to work with)
for row in db.execute("SELECT * FROM any_table LIMIT 10"): print(dict(row)) # rows (dictionaries)
for row in db.execute("select name from sqlite_master where type='table'").fetchall(): pprint.pprint(dict(row)) # tables (pprint)
for row in db.execute("SELECT * FROM stats LIMIT 10"): pprint.pprint((dict(row))) # rows (pprint)
''' Create '''
db.execute("CREATE TABLE events(id integer primary key, name text, start_date text, end_date text, description text);")

Basic Python

# Shuffle data
import numpy as np
data = np.arange(10)

indicies_permutation = np.random.permutation(len(data))
shuffled_data = data[indicies_permutation]

Tensorflow

Utilities

• One Hot Encode

from tensorflow.keras.utils import to_categorical
y_train = to_categorical(train_labels)

Models

1 Architecture model.Sequential([layers.Dense(16,actyivation="relu"),....])

• Simple stacked layers w/ relu activation can solve lots of problems
• Make last layer layers.Dense(1,activation="sigmoid") for binary classification
• Make last layer layers.Dense(#_categories, activation="softmax" for single label multiclassification

2 Compile model.compile(optimizer=_, loss=_,metrics=[_])

• Set Optimizer to "rmsprop" for 95% of models
• if binary classificationWhen then loss="binary_crossentropy"
• if single label multiclassification with OHE y then loss="categorical_crossentropy"
• if single label multiclassification with int y then loss="sparse_categorical_crossentropy"

3 Fit model.fit(x_train, y_train, epochs=_, batch_size=_2^n_)

• epochs is number of forward + backward pass the model makes attempting to reduce loss carefull not to overfit

NbDev

nbdev_new
nbdev_preview
nbdev_prepare
nbdev_export
# nbdev_watch... 

GIT

Basics

# Verify installed
git --version 
# Set up
git config --global user.name "Your Name"
git config --global user.email "[email protected]"

# Confirm
git config --list

# navigate to location
git init
ls -la # to verify
rm -rf .git # to stop tracking project with .git
git status # to see current status
touch .gitignore # add file names to ignore (can also use wildcards ex *.txt)

git add "file_name" # add a single file to staging area
git add -A # add all files to Staging Area
git reset # Remove files from the Staging Area
git status # to verify   

git add -A
git commit -m "Initial Commit" # try to make better Messages
git log # will show the History

Remote Repo

git clone <url> <where to clone> # will also work for local files
git remote -v # Lists Repo information
git branch -a # List all of the Branches of the Repo
# ..make some changes to code...
git diff
git status
git add -A
git commit -m "message"
# When ready to commit changes do a git pull -> git push (to capture any other changes)
git pull origin master
git push origin master

Branching

git branch "new_feature" # Makes the branch
git checkout "new_feature" # Moves you over the the branch
git add -A
git commit -m "small_new_feature" # uploads teh change ONLY to the branch
git push -u origin "new_feature" # slightly complex command that associates the brance with main
git branch -a # To verify
git add -A && git commit -m "small new feature..."
git push
# Ready to merge with Master Branch
git checkout master
git pull origin master
git merge "new_feature" # Merges the changes 
git push origin master # Uploads the changes to remote repo
#Delete the Branch
git branch -d "new_feature" # will delete localy
git push origin --delte "new_feature" # will delete from remote repo
git branch -s # to verify
# To go back to an old version and branch off
git log --graph --oneline --all --decorate # to see all committs # 
git checkout -b new-branch-name 4f8d768 # Creating a new branch from an old commit

Rolling back to a previous commit

Step 1: Reset to the desired commit

git init && git add -A && git commit -m "Initial commit"
git reset --hard origin/main # Roll back to last commit
git log --oneline --graph --decorate --all                  # With Page
git --no-pager log --oneline --graph --decorate --all       # No page

Useful Terminal (Linux/Mac)

• show ssh keys: ls -al ~/.ssh
• Copy: pbcopy < filename.txt
• Copy output: ls | pbcopy

SSH

1. create new ssh

ssh-keygen -t rsa -b 4096 -f ~/.ssh/my_server_key

2. Send ssh to server

ssh-copy-id -i ~/.ssh/custom_key_name.pub user@server_ip

3. Edit ~/.ssh/config

Host orange
    HostName server_ip
    User user   # same as step 2
    IdentityFile ~/.ssh/my_server_key
'''
>> ssh my_server_key # Will log you on
'''