100DaysOfMLCode

The creation of this repository was inspired by Siraj Raval's challenge to code machine learning for at least an hour everyday for 100 days.

I nervously accepted this challenge in addition to working full time and taking 6 hours of graduate courseowrk in the 2018 summer semester. I will use this repository to store code, jupyter notebook examples, and thought processes.

Topics Explored:

Day 1 - July 7 | Principal Component Analysis (PCA) and explained variance ratio

Day 2 - July 8 | SparsePCA -> CODE

Day 3 - July 9 | Bag of Words

Day 4 - July 10 | Tokenization & Vectorization time trials -> CODE

Day 5 - July 11 | Stemming and Lemmatizing with CountVectorizer, TfidfVectorizer, and HashingVectorizer -> CODE

Day 6 - July 12 | Development of visualization pipeline for ML -> CODE

Day 7 - July 13 | Big Data Visualization with Datashader

Day 8 - July 14 | t-SNE and Datashader Failure -> CODE

Day 9 - July 15 | Gene Expression - Getting Started -> FOLDER

Day 10 - July 16 | Gene Expression - Reading in Data

Day 11 - July 17 | Gene Expression - Preprocessing & Boxplot

Day 12 - July 18 | Intro to Data Splitting -> CODE

Day 13 - July 19 | Text Relationships with spaCy -> CODE

Day 14 - July 20 | Gene Expression - Cytoscape and Orange3

Day 15 - July 21 | Trial-and-error Data Splitting Research

Day 16 - July 22 | Trial-and-error Data Splitting Implimentation -> CODE

Day 17 - July 23 | NMF -> CODE

Day 18 - July 24 | RFE -> CODE

Day 19 - July 25 | Exploring Variable Replacement

Day 20 - July 26 | Pipelines - Introduction

Day 21 - July 27 | A list of 10,000 dictionaries -> CODE

Day 22 - July 28 | Linear Regression - Simple in R -> Folder

Day 23 - July 29 | Data Visualization, Dimensionality Reduction, Feature Selection, and a hand full of models. -> CODE

Day 24 - July 30 | Linear Regression - Continue to draft description -> Folder

Day 25 - July 31 | Linear Regression - Simple in Python -> CODE

Day 26 - Aug 1 | Pipeline - Start of Pipeline Example -> CODE

Day 27 - Aug 2 | Pipeline - Ridge Regression for Pipeline Example -> CODE

Day 28 - Aug 3 | Pipeline - Flexibility for selecting columns with missing values -> CODE

Day 29 - Aug 4 | Pipeline - Pipeline to compare methods of handling missing values -> CODE

Day 30 - Aug 5 | Pipeline - Identify categorical columns and convert to dummy -> CODE

Day 31 - Aug 6 | Pipeline - Custom Imputer using sklearn linear_model -> CODE

Day 32 - Aug 7 | kNN - add to Pipeline & normalizing -> CODE

Day 33 - Aug 8 | Pipeline - Researching topics to come

Day 34 - Aug 9 | What's great about bias?

Day 35 - Aug 10 | Bias-Variance decomposition - rounding error & elimination

Day 36 - Aug 11 | Bias-Variance decomposition from scratch in Python

Day 37 - Aug 12 | Continued work on Bias-Variance decomposition

Day 38 - Aug 13 | Bias-Variance decomposition working example

Day 39 - Aug 14 | Scatterplots for Collinearity

Day 40 - Aug 15 | ML Work for Client - not shared publicly

Day 41 - Aug 16 | Correlation Matrix for Collinearity

Day 42 - Aug 17 | Ontology from web scraping

Day 43 - Aug 18 | Eigen Values for MultiCollinearity

Day 44 - Aug 19 | Eigen Values & Vectors for MultiCollinearity

Day 45 - Aug 20 | Word frequencies from PDFs

Day 46 - Aug 21 | NLP with Regression - Expoloring the literature

Day 47 - Aug 22 | Text mining for Google Chips

Day 48 - Aug 23 | Methods of Web scraping

Day 49 - Aug 24 | Selenium for web scraping

Day 50 - Aug 25 | Reformatting results of web scraping

Day 51 - Aug 26 | NLP methods from web scraped results

Day 52 - Aug 27 | Applied Algorithms - different methods of sorting

Day 53 - Aug 28 | Methods of NLP for Social Media Data

List of Topics to Explore:

• PCA on Genetic Data - Gene Expression
  • Create Jupyter Notebook foundation
  • Find Good Data
  • Explain how to differentiate good data from bad data
• GPU
• Efficient Use of Data Structures
• Write computationally expensive parts in C++
• Make good use of memory & caching
• Multireading / multiprocessing in Python, Celery for parallel processing
• Kernal PCA
• Differences (pro/cons) between Stemming and Lemmatizing methods
• PCA to display failure risk
  • Lots / batches that take too long
  • Determine coorinary value
  • adjust threshold & critical thresholds
• Producing Production Quality code
• How tokenized data is used for ML algorithms
• Use of predeveloped vocabularies
  • i2b2
• Hypertools
  • Visualizing high dimensional data: https://hypertools.readthedocs.io/en/latest/
• MongoDB with Neo4j and Orient
• AutoML