One Hour Challenges

Quick burst challenge projects

Different projects I attempt with a one hour time limit. I'll set either a goal or library/methodology I want to use and attempt to complete the project within an hour.

ONE - Semi-Supervised Learning

GOAL: Can I cluster people by their demographic statistics and predict occupation?

METHOD: 
	- I use the audit data provided by Rattle to pull demographic and occupation data
	- Quick cleaning of the data into form suited for models to be used
	- Cluster the data using k-means 
	- Train a model with clusters as target dependent
	- Classify using k-nn to predict cluster
	- Does cluster relate to occupation at all?
	
RESULTS: 
	- Clustering results display defined segmentation between the give inputs (age, income, marital, gender).
	- Particularly defined groups at either k = 3.
		-> clusters = (young and high income, low income and male, older and high income)
	- Using a k = 5 results in a great predictor for cluster classification
		-> 100% accuracy predicting cluster in test data. (too accurate? why?)
	- Clusters very weakly relate to occupations. Not a success
	- Clusters do strongly correlate to age and income though 

OVERALL: Semi-supervised learning via clustering demographic data fails to predict occupation.

TIME TAKEN: 62 minutes

TWO - EDA w/ Matplotlib and Seaborn

GOAL: How many insights can I discover from an hour of eda with matplotlib/seaborn?

METHOD: 
	- Data obtained from Kaggle dataset: 'Pokemon with Stats'
	- I typically run eda in R so wanted to get practice in Py
	- Pull in data and drop incomplete rows
	- Get descriptive stats of data
	- Run univariate analysis of data
	- Run bivariate analysis of data
	- Dive deeper into initial insights with multivariate analysis.
	- What are the most interesting trends to visualize more deeply?
	
RESULTS: 
	- Learned how to subplot in matplotlib
	- Learned about querying in pandas
	- Linear Relationships between most card stats
	- Indicating different power levels rather than trade-offs between stats
	- Water type seems to be on the fringes suggesting more volatile stats exchanges
	- Grass types seem to be more consistent yet average
	- Stat totals about evenly distributed between generations

OVERALL: Seaborn is a great tool for making quick and beautiful charts when conducting EDA.

TIME TAKEN: 76 minutes

THREE - Recommender System using TF-IDF

GOAL: Can I suggest similar wines given an input wine?

METHOD: 
	- Data obtained from Kaggle dataset: 'Wine Reviews'
	- Get some practice with NLP and create a recommender system
	- Pull in data and drop incomplete rows
	- Text preprocessing
	- TF_IDF algorithm
	- Cosine distance similarity function
	- Suggest top 10 based on input
	
RESULTS: 
	- Recommendation wines that seem appropriate!
	- Description or original input and suggested are very similar
	- Indicting a Rec Sys which is working as expected.

OVERALL: Scikit-Learn's TF-IDF Vectorizer makes forming a blunt wine Rec Sys easy to do on tasting notes.

TIME TAKEN: 95 mins

FOUR - Recommender System and Similarity Visualization with Word2Vec

GOAL: Can I map out similar wines in a vector space to identify similar tastes and recommend new wines with? 

METHOD: 
	- Data obtained from Kaggle dataset: 'Wine Reviews'
	- Get some practice with NLP and learn to apply word2vec 
	- Pull in data and drop incomplete rows
	- Text preprocessing
	- Apply Word2Vec
	- Map out in vector space
	- Avg out wine description to place them on vector space
	- take given description input and recommend a wine based on it
	
RESULTS: 
	- word2vec is a powerful tool for identifying similarity between words/objects
	- was able to create 2-D matrix using cosine similarity to group similar wines based on description
	- User can input tasting notes they are looking for and code returns a list of top 5 similar tasting wines!

OVERALL: Word2Vec can create a very powerful and scalable recommendation system. 

TIME TAKEN: 180ish

FIVE - Mass Shooting Data Deep Dive w/ Tableau

GOAL: Identify Trends i= Data and Create Story-telling Visual?

METHOD: 
	- Data obtained from Mother Jones' 'Mass Shooting Investigation'
	- Pull in to Tableau
	- EDA
	- Create metrics?
	- Deep Dive Shooting trends
	- Create a Informative dashboard
	- Final Visualization Focus Identified and Created
	
RESULTS: 
	- Visualized a handful of informative charts on mass shootings, NICS background checks, and gun type relationships.

OVERALL: Standard fare Tableau exploration and visualization. Love the tool. This was a tough subject.

TIME TAKEN: 120ish