examples/movies/kms_with_aws_movie.md
In rdrr1990/kerasformula: A High-Level R Interface for Neural Nets
kerasformula: classification with AWS movie data
Pete Mohanty
AWS Movie Data with kerasformula
This document shows how to fit a neural net with kerasformula using an Amazon AWS database of about 3,000 popular movies.
library(kerasformula)
library(ggplot2)
movies <- read.csv("http://s3.amazonaws.com/dcwoods2717/movies.csv")
dplyr::glimpse(movies)
Observations: 2,961
Variables: 11
$ title <fct> Over the Hill to the Poorhouse, The Broadw...
$ genre <fct> Crime, Musical, Comedy, Comedy, Comedy, An...
$ director <fct> Harry F. Millarde, Harry Beaumont, Lloyd B...
$ year <int> 1920, 1929, 1933, 1935, 1936, 1937, 1939, ...
$ duration <int> 110, 100, 89, 81, 87, 83, 102, 226, 88, 14...
$ gross <int> 3000000, 2808000, 2300000, 3000000, 163245...
$ budget <int> 100000, 379000, 439000, 609000, 1500000, 2...
$ cast_facebook_likes <int> 4, 109, 995, 824, 352, 229, 2509, 1862, 11...
$ votes <int> 5, 4546, 7921, 13269, 143086, 133348, 2918...
$ reviews <int> 2, 107, 162, 164, 331, 349, 746, 863, 252,...
$ rating <dbl> 4.8, 6.3, 7.7, 7.8, 8.6, 7.7, 8.1, 8.2, 7....
How the data are cleaned affects overfitting (models that do relatively well on training data compared to test data). The first model omits director, the second includes, and the third includes dummies for top director (by frequency of appearance in the data) and codes the rest as "other". In general with neural nets, overfitting is the big concern. Here, we only have a relatively small number of columns and so the regularizers are disabled (as opposed to their default, L1).
head(sort(table(movies$director), decreasing = TRUE))
Steven Spielberg Clint Eastwood Martin Scorsese Tim Burton
23 19 16 16
Spike Lee Steven Soderbergh
15 15
sum(table(movies$director) == 1) # number of directors appearing only once
[1] 788
Classifying Genre
sort(table(movies$genre))
Thriller Musical Romance Western Family Sci-Fi
1 2 2 2 3 7
Mystery Documentary Fantasy Animation Horror Biography
16 25 28 35 131 135
Crime Adventure Drama Action Comedy
202 288 498 738 848
out1 <- kms(genre ~ . -director -title, movies, seed = 12345, activation = "relu", use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 2048
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 102,528
Trainable params: 102,528
Non-trainable params: 0
___________________________________________________________________________
plot(out1$history) + labs(title = "Classifying Genre",
subtitle = "Source data: http://s3.amazonaws.com/dcwoods2717/movies.csv", y="") + theme_minimal()
The model is only able to classify about half the movies by genre correctly. Does adding director help?
out2 <- kms(genre ~ . -title, movies, activation = "relu", seed = 12345, use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 351488
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 451,968
Trainable params: 451,968
Non-trainable params: 0
___________________________________________________________________________
Doesn't hurt much but introduces overfitting.... Including only the top directors doesn't make big improvements but doesn't have the overfitting issue.
movies$top50_director <- as.character(movies$director)
movies$top50_director[rank(movies$director) > 50] <- "other"
out3 <- kms(genre ~ . -director -title, movies, activation = "relu", seed = 12345, use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 9216
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 109,696
Trainable params: 109,696
Non-trainable params: 0
___________________________________________________________________________
What's going on? The above plots show epoch-by-epoch training data vis-a-vis validation data. By default, kms holds out an additional 20% of the data as test data. That test data is used to construct a confusion matrix, or, when there are many categories, a table that reports the proportion correct as well as most common errors (?confusion for options). For example, comedies are most likely to be mistaken for dramas and actions for comedies. Biographies, a relatively rare category, are usually mistakan for dramas.
out1$confusion
Action_pred Adventure_pred Animation_pred Biography_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Comedy_pred Crime_pred Documentary_pred Drama_pred
Action_obs 148 0 0 0
Adventure_obs 60 0 0 0
Animation_obs 9 0 0 0
Biography_obs 24 0 0 0
Comedy_obs 176 0 0 0
Crime_obs 44 0 0 0
Documentary_obs 5 0 0 0
Drama_obs 105 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 7 0 0 0
Horror_obs 28 0 0 0
Musical_obs 1 0 0 0
Mystery_obs 4 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 1 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Family_pred Fantasy_pred Horror_pred Musical_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Mystery_pred Romance_pred Sci-Fi_pred Thriller_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Western_pred
Action_obs 0
Adventure_obs 0
Animation_obs 0
Biography_obs 0
Comedy_obs 0
Crime_obs 0
Documentary_obs 0
Drama_obs 0
Family_obs 0
Fantasy_obs 0
Horror_obs 0
Musical_obs 0
Mystery_obs 0
Romance_obs 0
Sci-Fi_obs 0
Thriller_obs 0
Western_obs 0
That means we can visualize the models' confusion like so:
plot_confusion(model1, model2, model3, title="Classifying Genre", subtitle="Source Data: http://s3.amazonaws.com/dcwoods2717/movies.csv")
The real choice appears to be between Model 1 and Model 3 with perhaps a faint edge to Model 1. batch_size was set to 1 to give the estimator more of fighting chance for rare outcomes. For a more general introduction to that shows how to change loss, layer type and number, activation, etc. see package vignettes or this example using Twitter data.
rdrr1990/kerasformula documentation built on June 6, 2019, 8:02 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
kerasformula: classification with AWS movie data
Pete Mohanty
AWS Movie Data with kerasformula
This document shows how to fit a neural net with kerasformula using an Amazon AWS database of about 3,000 popular movies.
library(kerasformula)
library(ggplot2)
movies <- read.csv("http://s3.amazonaws.com/dcwoods2717/movies.csv")
dplyr::glimpse(movies)
Observations: 2,961
Variables: 11
$ title <fct> Over the Hill to the Poorhouse, The Broadw...
$ genre <fct> Crime, Musical, Comedy, Comedy, Comedy, An...
$ director <fct> Harry F. Millarde, Harry Beaumont, Lloyd B...
$ year <int> 1920, 1929, 1933, 1935, 1936, 1937, 1939, ...
$ duration <int> 110, 100, 89, 81, 87, 83, 102, 226, 88, 14...
$ gross <int> 3000000, 2808000, 2300000, 3000000, 163245...
$ budget <int> 100000, 379000, 439000, 609000, 1500000, 2...
$ cast_facebook_likes <int> 4, 109, 995, 824, 352, 229, 2509, 1862, 11...
$ votes <int> 5, 4546, 7921, 13269, 143086, 133348, 2918...
$ reviews <int> 2, 107, 162, 164, 331, 349, 746, 863, 252,...
$ rating <dbl> 4.8, 6.3, 7.7, 7.8, 8.6, 7.7, 8.1, 8.2, 7....
How the data are cleaned affects overfitting (models that do relatively well on training data compared to test data). The first model omits director, the second includes, and the third includes dummies for top director (by frequency of appearance in the data) and codes the rest as "other". In general with neural nets, overfitting is the big concern. Here, we only have a relatively small number of columns and so the regularizers are disabled (as opposed to their default, L1).
head(sort(table(movies$director), decreasing = TRUE))
Steven Spielberg Clint Eastwood Martin Scorsese Tim Burton
23 19 16 16
Spike Lee Steven Soderbergh
15 15
sum(table(movies$director) == 1) # number of directors appearing only once
[1] 788
Classifying Genre
sort(table(movies$genre))
Thriller Musical Romance Western Family Sci-Fi
1 2 2 2 3 7
Mystery Documentary Fantasy Animation Horror Biography
16 25 28 35 131 135
Crime Adventure Drama Action Comedy
202 288 498 738 848
out1 <- kms(genre ~ . -director -title, movies, seed = 12345, activation = "relu", use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 2048
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 102,528
Trainable params: 102,528
Non-trainable params: 0
___________________________________________________________________________
plot(out1$history) + labs(title = "Classifying Genre",
subtitle = "Source data: http://s3.amazonaws.com/dcwoods2717/movies.csv", y="") + theme_minimal()
The model is only able to classify about half the movies by genre correctly. Does adding director help?
out2 <- kms(genre ~ . -title, movies, activation = "relu", seed = 12345, use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 351488
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 451,968
Trainable params: 451,968
Non-trainable params: 0
___________________________________________________________________________
Doesn't hurt much but introduces overfitting.... Including only the top directors doesn't make big improvements but doesn't have the overfitting issue.
movies$top50_director <- as.character(movies$director)
movies$top50_director[rank(movies$director) > 50] <- "other"
out3 <- kms(genre ~ . -director -title, movies, activation = "relu", seed = 12345, use_bias = FALSE, kernel_regularizer = NULL, bias_regularizer = NULL, activity_regularizer = NULL)
units activation dropout use_bias kernel_initializer embedding
layer1 256 relu 0.4 FALSE glorot_uniform FALSE
layer2 128 relu 0.4 FALSE glorot_uniform NA
layer3 17 relu 0.0 FALSE glorot_uniform NA
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
dense_1 (Dense) (None, 256) 9216
___________________________________________________________________________
dense_2 (Dense) (None, 256) 65536
___________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
___________________________________________________________________________
dense_3 (Dense) (None, 128) 32768
___________________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
___________________________________________________________________________
dense_4 (Dense) (None, 17) 2176
===========================================================================
Total params: 109,696
Trainable params: 109,696
Non-trainable params: 0
___________________________________________________________________________
What's going on? The above plots show epoch-by-epoch training data vis-a-vis validation data. By default, kms holds out an additional 20% of the data as test data. That test data is used to construct a confusion matrix, or, when there are many categories, a table that reports the proportion correct as well as most common errors (?confusion for options). For example, comedies are most likely to be mistaken for dramas and actions for comedies. Biographies, a relatively rare category, are usually mistakan for dramas.
out1$confusion
Action_pred Adventure_pred Animation_pred Biography_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Comedy_pred Crime_pred Documentary_pred Drama_pred
Action_obs 148 0 0 0
Adventure_obs 60 0 0 0
Animation_obs 9 0 0 0
Biography_obs 24 0 0 0
Comedy_obs 176 0 0 0
Crime_obs 44 0 0 0
Documentary_obs 5 0 0 0
Drama_obs 105 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 7 0 0 0
Horror_obs 28 0 0 0
Musical_obs 1 0 0 0
Mystery_obs 4 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 1 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Family_pred Fantasy_pred Horror_pred Musical_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Mystery_pred Romance_pred Sci-Fi_pred Thriller_pred
Action_obs 0 0 0 0
Adventure_obs 0 0 0 0
Animation_obs 0 0 0 0
Biography_obs 0 0 0 0
Comedy_obs 0 0 0 0
Crime_obs 0 0 0 0
Documentary_obs 0 0 0 0
Drama_obs 0 0 0 0
Family_obs 0 0 0 0
Fantasy_obs 0 0 0 0
Horror_obs 0 0 0 0
Musical_obs 0 0 0 0
Mystery_obs 0 0 0 0
Romance_obs 0 0 0 0
Sci-Fi_obs 0 0 0 0
Thriller_obs 0 0 0 0
Western_obs 0 0 0 0
Western_pred
Action_obs 0
Adventure_obs 0
Animation_obs 0
Biography_obs 0
Comedy_obs 0
Crime_obs 0
Documentary_obs 0
Drama_obs 0
Family_obs 0
Fantasy_obs 0
Horror_obs 0
Musical_obs 0
Mystery_obs 0
Romance_obs 0
Sci-Fi_obs 0
Thriller_obs 0
Western_obs 0
That means we can visualize the models' confusion like so:
plot_confusion(model1, model2, model3, title="Classifying Genre", subtitle="Source Data: http://s3.amazonaws.com/dcwoods2717/movies.csv")
The real choice appears to be between Model 1 and Model 3 with perhaps a faint edge to Model 1. batch_size was set to 1 to give the estimator more of fighting chance for rare outcomes. For a more general introduction to that shows how to change loss, layer type and number, activation, etc. see package vignettes or this example using Twitter data.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.