AI summary by TwinMind
Three (3) things to do when starting out in Data Science
StatQuest with Josh Starmer
11m 10s
Introduction & Overview
- Josh Starmer presents three essential things for data science beginners
- Focus on curiosity as core principle across all aspects
Thing 1: Be Curious About the Data
- Data consists of numbers/words in spreadsheets but has deeper story
- Data collection involves people at every step, creating complexity beyond raw numbers
Understanding Data Sources & Context
- Examples: clinical trials, research labs, website interactions, weather sensors, surveys, political polls
- Context crucial for interpretation - need to know about data subjects
- Demographics (age, race, ethnicity)
- Behaviors (smoking status)
- These factors limit what can be extrapolated from results
Bias in Data Collection
- Example: lung cancer study with only white males
- Results may not apply to majority population
- Importance of representative sampling
Data Collection Process Impact
- Cell extraction expertise level affects sample purity/contamination
- Expert vs first-time extractor creates different confidence levels
- Data preparation expertise affects final dataset accuracy
- Inexperienced preparation can corrupt analysis inputs
Recommendations for Data Understanding
- Meet patients/subjects directly - may notice critical details others miss
- Observe data collection processes firsthand
- Meet dataset creators to validate assumptions about data
- Data science = "people science" - understanding people as important as understanding data
Thing 2: Be Curious About the Methods
- Data science field full of methods: statistics, machine learning, databases
- Blessing (many tools) but also curse (time-consuming to master)
- Critical to understand method limitations
Neural Network Example - Drug Dosage Study
- Simple dataset: low/high dosages ineffective, medium dosage effective
- Neural network produced "green squiggle" that fit training data but had weird characteristics
- Predictions went above 1 when training data never exceeded 1
- Created unexpected "bump" between data points
Method Limitations Revealed
- Neural networks only penalized where training data exists
- Can behave unpredictably between data points
- Could produce bigger bumps, multiple bumps, or other unexpected patterns
- Human intuition would fit smooth arc, but neural network fits whatever minimizes training error
Real-world Implications
- Autonomous vehicle example: humans extrapolate between experiences predictably
- Neural networks may behave unpredictably between training scenarios
- Other drivers may not relate to neural network decisions
- Understanding method behavior crucial for trust and safety
Thing 3: Just Be Curious (About the Big Picture)
- Data science work is small but important piece of larger puzzle
- Not just "plugging in data and hitting return"
Broader Impact Areas
- Helping cure diseases
- Lifting people out of poverty
- Bringing smiles to faces
- Helping people in new/exciting ways
Ethical Considerations
- Work always serves people, not isolated in cave
- Must be aware of how work is used
- Question: Is work making world better place?
- Only way to know: be curious about other puzzle pieces
- Must see big picture yourself - can't be shown by others
Conclusion
- Curiosity as fundamental principle across all three areas
- Encouragement to subscribe, support via Patreon, channel membership, merchandise, donations
