Deep learning in machine learning

Deep learning is a subject of importance right from image detection to speech recognition and AI-related activity

> library(neuralnet)
> library(ISLR)
Error in library(ISLR) : there is no package called ‘ISLR’
> install.packages("ISLR")
> library(ISLR)
Warning message:
package ‘ISLR’ was built under R version 3.5.2
> library(ISLR)
> data = College

Image detection

> View(data)
>
> max_data <- apply(data[,2:18], 2, max)
> min_data <- apply(data[,2:18], 2, min)
> data_scaled <- scale(data[,2:18],center = min_data, scale = max_data - min_data)
>
> Private = as.numeric(College$Private)-1
> data_scaled = cbind(Private,data_scaled)
> index = sample(1:nrow(data),round(0.70*nrow(data)))
>
> train_data <- as.data.frame(data_scaled[index,])
> test_data <- as.data.frame(data_scaled[-index,])
>
> n = names(train_data)
>
> f <- as.formula(paste("Private ~", paste(n[!n %in% "Private"], collapse = " + ")))
>
> deep_net = neuralnet(f,data=train_data,hidden=c(5,3),linear.output=F)
> plot(deep_net)

> predicted_data <- compute(deep_net,test_data[,2:18])
>
> print(head(predicted_data$net.result))
                          [,1]
Alaska Pacific University    1
Albertson College            1
Allegheny College            1
Amherst College              1
Andrews University           1
Antioch University           1
> predicted_data$net.result <- sapply(predicted_data$net.result,round,digits=0)
> table(test_data$Private,predicted_data$net.result)
 
      0   1
  0  54  10
  1  10 159
> install.packages("h2o")
Installing package into ‘C:/Users/ADMIN/Documents/R/win-library/3.5’
(as ‘lib’ is unspecified)
trying URL 'https://cran.rstudio.com/bin/windows/contrib/3.5/h2o_3.22.1.1.zip'
Content type 'application/zip' length 122132920 bytes (116.5 MB)
downloaded 116.5 MB

package ‘h2o’ successfully unpacked and MD5 sums checked

The downloaded binary packages are in
C:\Users\ADMIN\AppData\Local\Temp\Rtmpof4ihy\downloaded_packages
> library(h2o)

----------------------------------------------------------------------

Your next step is to start H2O:
    > h2o.init()

For H2O package documentation, ask for help:
    > ??h2o

After starting H2O, you can use the Web UI at http://localhost:54321
For more information visit http://docs.h2o.ai

----------------------------------------------------------------------


Attaching package: ‘h2o’

The following objects are masked from ‘package:stats’:

    cor, sd, var

The following objects are masked from ‘package:base’:

    %*%, %in%, &&, ||, apply, as.factor,
    as.numeric, colnames, colnames<-, ifelse,
    is.character, is.factor, is.numeric, log,
    log10, log1p, log2, round, signif, trunc

Warning message:
package ‘h2o’ was built under R version 3.5.2
> library(h2o)
> c1=h2o.init(max_mem_size = "2G",
+             nthreads = 2,
+             ip = "localhost",
+             port = 54321)

H2O is not running yet, starting it now...

Note:  In case of errors look at the following log files:
    C:\Users\ADMIN\AppData\Local\Temp\Rtmpof4ihy/h2o_ADMIN_started_from_r.out
    C:\Users\ADMIN\AppData\Local\Temp\Rtmpof4ihy/h2o_ADMIN_started_from_r.err

java version "1.8.0_151"
Java(TM) SE Runtime Environment (build 1.8.0_151-b12)
Java HotSpot(TM) 64-Bit Server VM (build 25.151-b12, mixed mode)

Starting H2O JVM and connecting: .. Connection successful!

R is connected to the H2O cluster:
    H2O cluster uptime:         8 seconds 865 milliseconds
    H2O cluster timezone:       Asia/Kolkata
    H2O data parsing timezone:  UTC
    H2O cluster version:        3.22.1.1
    H2O cluster version age:    2 months and 1 day
    H2O cluster name:           H2O_started_from_R_ADMIN_unb933
    H2O cluster total nodes:    1
    H2O cluster total memory:   1.78 GB
    H2O cluster total cores:    2
    H2O cluster allowed cores:  2
    H2O cluster healthy:        TRUE
    H2O Connection ip:          localhost
    H2O Connection port:        54321
    H2O Connection proxy:       NA
    H2O Internal Security:      FALSE
    H2O API Extensions:         Algos, AutoML, Core V3, Core V4
    R Version:                  R version 3.5.0 (2018-04-23)

> data(iris)
> summary(iris)
  Sepal.Length       Sepal.Width        Petal.Length
 Min.   :4.300000   Min.   :2.000000   Min.   :1.000
 1st Qu.:5.100000   1st Qu.:2.800000   1st Qu.:1.600
 Median :5.800000   Median :3.000000   Median :4.350
 Mean   :5.843333   Mean   :3.057333   Mean   :3.758
 3rd Qu.:6.400000   3rd Qu.:3.300000   3rd Qu.:5.100
 Max.   :7.900000   Max.   :4.400000   Max.   :6.900
  Petal.Width             Species
 Min.   :0.100000   setosa    :50
 1st Qu.:0.300000   versicolor:50
 Median :1.300000   virginica :50
 Mean   :1.199333               
 3rd Qu.:1.800000               
 Max.   :2.500000               
> iris_d1 <- h2o.deeplearning(1:4,5,
+                             as.h2o(iris),hidden=c(5,5),
+                             export_weights_and_biases=T)
  |==============================================| 100%
  |==============================================| 100%
>
> iris_d1
Model Details:
==============

H2OMultinomialModel: deeplearning
Model ID:  DeepLearning_model_R_1551499097956_1
Status of Neuron Layers: predicting Species, 3-class classification, multinomial distribution, CrossEntropy loss, 73 weights/biases, 4.6 KB, 1,500 training samples, mini-batch size 1
  layer units      type dropout       l1       l2
1     1     4     Input  0.00 %       NA       NA
2     2     5 Rectifier  0.00 % 0.000000 0.000000
3     3     5 Rectifier  0.00 % 0.000000 0.000000
4     4     3   Softmax      NA 0.000000 0.000000
  mean_rate rate_rms momentum mean_weight weight_rms
1        NA       NA       NA          NA         NA
2  0.003087 0.003858 0.000000    0.055145   0.563931
3  0.001105 0.000828 0.000000    0.148419   0.333557
4  0.001324 0.000642 0.000000   -0.028787   1.654691
  mean_bias bias_rms
1        NA       NA
2  0.852716 0.826724
3  1.062501 0.273368
4  0.001277 0.103763


H2OMultinomialMetrics: deeplearning
** Reported on training data. **
** Metrics reported on full training frame **

Training Set Metrics:
=====================

Extract training frame with `h2o.getFrame("iris_sid_a830_1")`
MSE: (Extract with `h2o.mse`) 0.06891309264
RMSE: (Extract with `h2o.rmse`) 0.2625130333
Logloss: (Extract with `h2o.logloss`) 0.2637877632
Mean Per-Class Error: 0.07333333333
Confusion Matrix: Extract with `h2o.confusionMatrix(<model>,train = TRUE)`)
=========================================================================
Confusion Matrix: Row labels: Actual class; Column labels: Predicted class
           setosa versicolor virginica  Error
setosa         50          0         0 0.0000
versicolor      7         40         3 0.2000
virginica       0          1        49 0.0200
Totals         57         41        52 0.0733
                 Rate
setosa     =   0 / 50
versicolor =  10 / 50
virginica  =   1 / 50
Totals     = 11 / 150

Hit Ratio Table: Extract with `h2o.hit_ratio_table(<model>,train = TRUE)`
=======================================================================
Top-3 Hit Ratios:
  k hit_ratio
1 1  0.926667
2 2  1.000000
3 3  1.000000
> plot(iris_d1)
>
> h2o.weights(iris_d1, matrix_id=1)
   Sepal.Length    Sepal.Width  Petal.Length
1  0.2409946620  0.35679787397 -1.1170697212
2  0.6530707479  0.67512649298  0.2999281287
3 -0.1507256031  0.44363185763  0.5350508094
4  0.6421241760  0.03761408478 -0.1086364165
5 -0.2883346975 -0.27492222190  0.1764629483
    Petal.Width
1 -1.4844481945
2  0.2284315377
3  0.7389914393
4  0.0453807041
5 -0.5465673208

[5 rows x 4 columns]
> h2o.weights(iris_d1, matrix_id=2)
             C1             C2            C3
1 -0.2234841734 -0.01124510914 0.43649563193
2 -0.1895816028  0.21325439215 0.79128360748
3 -0.3035679460 -0.55753493309 0.48393419385
4  0.1758618206 -0.35445415974 0.36051633954
5  0.3964880109  0.23581361771 0.04750946537
             C4             C5
1 0.58218020201  0.02807313576
2 0.08547680080 -0.19191685319
3 0.45826247334  0.19889642298
4 0.08671515435  0.68862718344
5 0.20328296721  0.06957834214

[5 rows x 5 columns]
> h2o.weights(iris_d1, matrix_id=3)
             C1           C2           C3            C4
1 -0.6639142036 -2.032027721 -2.748629332  0.7574009895
2 -0.6556437612 -0.368837446 -1.755167603  0.5674321651
3  2.5461874008 -1.720250487  2.794062376 -1.1157760620
            C5
1 1.7997418642
2 1.4643006325
3 0.6993095279

[3 rows x 5 columns]
> h2o.biases(iris_d1, vector_id=1)
            C1
1 2.4911530990
2 0.6355656164
3 0.4266453373
4 0.3115003118
5 0.3987136025

[5 rows x 1 column]
> h2o.biases(iris_d1, vector_id=2)
            C1
1 0.7540707757
2 1.1464952188
3 0.7208821239
4 1.2750280645
5 1.4160310664

[5 rows x 1 column]
> h2o.biases(iris_d1, vector_id=3)
              C1
1  0.09313390534
2  0.05446989759
3 -0.14377270651

[3 rows x 1 column]
>
> plot(as.data.frame(h2o.weights(iris_d1, matrix_id=1))[,1])