COMU1130 TUT 2/3 Limited data set of academies - High budget films - International productions - Unethical - took data - Only successful applications o Determining failures - Those that didn't win were still success - Criterion that values the outcome - IMDB rating is only targeted at certain audience o Not everyone is rating o Not everyone is visiting website IMBd rating chart (academy award films) - Highest score was low - Medium was a 7 Variables selected from IMBd - Language - Country - runtime - genre - director - writer - actors - production Feature engineering - Selected features need to be either o Numeric (real numbers) o Categorical (limited number of categories not of a high number) Dichotomous - they are, or they are not (1 or 0) Histogram - Steps Kim has done till now Collected data Identified outcome (IMBd rating) Identified sources for features (i.e., feature selection) - Genre - Country - Director Transformed raw feature data into manageable categories
NEEDS TO SELECT AND TRAIN A MODEL (this tutorial) - Develop formular Model selection - Linear regression: fully transparent - Support vector machine: relatively black-boxed Linear regression How would you describe the relationship between the duration of a project and its rating? - Y=rating =, x=duration (runtime) - As the length increases so does the rating < relationship - Positive correlation/ relationship o As one increases so does the other - Use a trendline/ line of best fit o Reflects general tendency of the relationship o Collapse's information in a simple generalised representation o Usually runs in the middle of data Y = a+(b * x) - Intercept o Value of y when x=0 o Slope is when x+1 - Gradient o B value o On average, per minute increase in duration, the rating goes up with .0837 units For other features Y = a + b (b1 * x1) + ... + (bn * Xn) Kim made poor choices in data collection and feature selection/ engineering and its starting to show ... Tut 3/3