Does having a weird seed mess up neural networks

nnet_seed.R

library(nnet)
library(dplyr)

set.seed(20200408)

diamonds <- diamonds %>% 
              select(carat, price, depth, table, x, y, z) %>% 
              scale
# Kinda dumb. I want reaosnable size datasets with no overlap
train_ii <- sample(nrow(diamonds), 1000)
test_ii <- sample(nrow(diamonds), 3000)
test_ii <- test_ii[!(test_ii %in% train_ii)]
train <- diamonds[train_ii, ]
test <- diamonds[test_ii, ]
test_y <- diamonds[, 'price'][test_ii]

mae <- rep(NA, 200)
for(i in 1:200){
  m <- nnet(price ~ depth + table + carat + x + y + z, data = train, size = 100, linout = TRUE)
  preds <- predict(m, newdata = test)
  mae[i] <- mean(abs(preds - test_y))
}

set.seed(2631)
m <- nnet(price ~ depth + table + carat + x + y + z, data = train, size = 100, linout = TRUE)
preds <- predict(m, newdata = test)
mae_odd <- mean(abs(preds - test_y))

hist(mae)
abline(v = mae_odd)