conormm

num_df <- num_df %>% 
  mutate(years_since_release = 2017 - title_year) %>% 
  select(-title_year)

conormm

df %>% 
  group_by(title_year) %>% 
  summarise(mean_rating = mean(imdb_score), 
            upper_rating = quantile(imdb_score, 0.975), 
            lower_rating = quantile(imdb_score, 0.0275)) %>% 
  ggplot(aes(title_year, mean_rating)) +
  geom_line(colour = "dodger blue") + 
  geom_point(alpha = 0.5) + 
  geom_smooth(method = "lm", colour = "red", alpha = 0.6, se = FALSE) +
  geom_errorbar(aes(ymin = upper_rating, ymax = lower_rating)) + 

conormm

correlations <- corrr::correlate(num_df) %>%
    gather(variable, correlation, 2:16) %>%
    select(rowname, variable, correlation) %>% 
    mutate(high_correlation = ifelse(abs(correlation) > 0.50, "high", "not so high"))

correlations %>% 
  ggplot(aes(reorder(rowname, correlation), reorder(variable, correlation), fill = correlation)) + 
  geom_tile(alpha = 0.6, colour = "black") +
  geom_text(aes(label = round(correlation, 2), colour = high_correlation)) +
  theme_minimal() +

conormm

fit <- num_df %>% 
  map(scale) %>%
  as.data.frame() %>% 
  bootstrap(10) %>% 
  do(tidy(lm(imdb_score ~., data = .), conf.int = TRUE)) %>% 
  mutate(sig_0005 = p.value < 0.005)

fit %>% 
  filter(term != "(Intercept)") %>% 
  group_by(term) %>% 

conormm

fit %>% 
  filter(term != "(Intercept)") %>% 
  group_by(term) %>% 
  summarise(estimate = mean(estimate), 
            conf.low = mean(conf.low), 
            conf.high = mean(conf.high), 
            sig_0005 = ifelse(sum(sig_0005) > 6, "significant (< 0.005)", "not significant")) %>% 
  ggplot(aes(y = reorder(term, estimate), x = estimate, colour = sig_0005)) +
  geom_point() + 
  geom_errorbarh(aes(xmax = conf.high, xmin = conf.low)) +

conormm

library(tidyverse)
library(here)
library(broom)
library(corrr)
library(forcats)
library(stringr)
library(lubridate)
library(gridExtra)

df <- read_csv("movie_metadata.csv")

conormm

library(dplyr)
library(ggplot2)

n     <- 200 # number of observations
bias  <- 4
slope <- 3.5
dot   <- `%*%` # defined for personal preference 

x   <- rnorm(n) * 2
x_b <- cbind(x, rep(1, n))

conormm

for (iteration in seq_len(n_iterations)) { 
  yhat               <- dot(x_b, theta)          # predict using weights in theta
  residuals_b        <- yhat - y                 # calculate the residuals
  gradients          <- 2/n * dot(t(x_b), residuals_b) # calculate the gradients of MSE w.r.t model weights 
  theta              <- theta - learning_rate * gradients # update theta 
  
  sse_i[[iteration]] <- sum((y - dot(x_b, theta))**2)
  b0[[iteration]]    <- theta[2]
  b1[[iteration]]    <- theta[1]
  

conormm

p1 <- df %>% 
  ggplot(aes(x=x, y=y)) + 
  geom_abline(aes(intercept = b0, 
                  slope = b1, 
                  colour = -sse, 
                  frame = model_iter), 
              data = model_i, 
              alpha = .50 
              ) +
  geom_point(alpha = 0.4) + 

conormm

predict_from_theta <- function(x) {
  
  x <- cbind(x, rep(1, length(x)))
  dot(x, theta)
  
}

predict_from_theta(rnorm(10))
           [,1]
 [1,] -1.530065