Aprendizaje de Máquina. Tarea 2: Gradient Descent and Learning Rates

Profesor: Felipe González

Elaborado por Daniela Pinto Veizaga

Parte 1: descenso en gradiente

Resolveremos un problema de minimización usando descenso en gradiente. Considera la siguiente función

library(tidyverse)

h <- function(x) {
  w <- x - 4
  (1/50) * (w^4 + w^2 + 100 * w)
}

Pregunta. grafica la función h. ¿Dónde está el mínimo? ¿Cuál es el valor que toma la función en el mínimo?

x <- c(-100:100)
h_x<- tibble(h(x),x)
h_x

h_x<- tibble(h(x),x)
ggplot(h_x) + geom_line(aes(x=x, y=h(x))) + labs(title = 'Función h')

print(paste0('El valor mínimo está en x=1, en ese punto, la función h_x() toma el siguiente valor: ', min(h_x$`h(x)`)))

[1] "El valor mínimo está en x=1, en ese punto, la función h_x() toma el siguiente valor: -4.2"

Pregunta: Ahora calcula la derivada en la siguiente función (rellena):

# calcula la derivada
h_deriv <- function(x) {
  # aquí tu calculo, debe regresar la derivada de h 
  # evaluada en x
  w <- x - 4
  deriv <- (1/25) * (2*(w**3) + w + 50)
  #deriv <- (4*(x-4)^3 + 2*(x-4) +100)*1/150
  deriv
}

Usaremos el código de la clase anterior:

# función para descenso
descenso <- function(n, z_0, eta, h_deriv){
  z <- matrix(0,n, length(z_0))
  z[1, ] <- z_0
  for(i in 1:(n-1)){
    # paso de descenso
    z[i+1, ] <- z[i, ] - eta * h_deriv(z[i, ])
  }
  z
}

Pregunta: Empezamos las iteraciones en 0. En el siguiente código, escoge un tamaño de paso (\(eta\)) demasiado grande (diverge), demasiado chico (tarda mucho en converger) y uno intermedio donde alcances el mínimo en un número de iteraciones razonable. Puede ser necesario que ajustes el número de iteraciones, y puede ser que obtengas desbordes si pones tamaños de paso demasiado grandes:

# pon un valor grande y uno chico para eta. ¿Qué pasa?
z_0 <- 0

for (i in c(0.001, 0.01, 0.1, 0.3, 0.345, 1)){
  eta <- i 
  name = paste('Learning rate',i,sep=" ")

  z <- descenso(100, z_0, eta, h_deriv)

  #Grafica las iteraciones
  dat_iteraciones <- tibble(iteracion = 1:nrow(z), 
                          x = z[, 1], y = h(z[, 1]))
  p<-ggplot(dat_iteraciones, aes(x = iteracion, y = x)) + geom_point() + geom_line()+ ggtitle(name)
  assign(paste0("plot", i), p)

}

library(cowplot)
plot_grid(plot0.001, plot0.01, plot0.1, plot0.3, plot0.345, plot1,
          labels = 'AUTO', label_fontfamily = "serif", label_fontface = "plain",
          label_colour = "black", ncol = 2)

Parte 2: descenso en gradiente para regresión

Pregunta: en el siguiente ejercicio ajustamos un modelo de regresión para el problema de precio de casas que vimos en clase. Escoge

1. Un número de iteraciones y tamaño de paso adecuados para encontrar los coeficientes,
2. ¿Qué pasa si pones un tamaño de paso muy chico? ¿Qué pasa si pones un tamaño de paso demasiado grande?

Cargamos los datos:

casas_receta <- read_rds("casas_receta.rds")
x_ent <- casas_receta %>% prep %>% juice %>% select(tiene_piso_2, calidad) %>% as.matrix()
y_ent <- casas_receta %>% prep %>% juice %>% pull(precio_m2_miles)

Pregunta: qué contienen x_ent y y_ent?

• x_ent contiene los features con los cuáles se hará la predicción; en este caso son dos features: tiene_piso_2 y calidad.

x_ent %>% summary

  tiene_piso_2       calidad  
 Min.   :0.0000   Min.   :-5  
 1st Qu.:0.0000   1st Qu.:-1  
 Median :0.0000   Median : 0  
 Mean   :0.4468   Mean   : 0  
 3rd Qu.:1.0000   3rd Qu.: 1  
 Max.   :1.0000   Max.   : 4  

• y_ent contiene la predicción, el precio por metro cuadrado.

x_ent y y_ent caracterizan a los datos de entrenamiento.

Definimos un modelo de regresión lineal en keras (nota: si quieres hacer corridas desde cero empieza con estas líneas que siguen. De otra manera vas a iterar desde el punto donde te hayas quedado en la corrida anterior):

library(tidyverse)
library(tidymodels)
library(keras)
 # ajusta este valor, es le tamaño de paso o tasa de aprendizaje

casas_receta <- read_rds("casas_receta.rds")
x_ent <- casas_receta %>% prep %>% juice %>% select(tiene_piso_2, calidad) %>% as.matrix()

y_ent <- casas_receta %>% prep %>% juice %>% pull(precio_m2_miles)
   
for (i in c(0.001, 0.01, 0.1, 0.2, 0.3)){   
   i= as.character(i)
   name = paste('logs/run_',i,sep="")
   n_entrena <- nrow(x_ent)
   modelo_casas <- keras_model_sequential() 
   modelo_casas %>%
        layer_dense(units = 1,activation = "linear",
         kernel_initializer = initializer_constant(0),
          bias_initializer = initializer_constant(0))   
   modelo_casas %>% compile(
       loss = "mean_squared_error",  # pérdida cuadrática
        optimizer = optimizer_sgd(lr = lr), # descenso en gradiente
        #metrics = list("mean_squared_error"),
        metrics = c('accuracy')
)
   
   historia <- modelo_casas %>% fit(
      as.matrix(x_ent), # x entradas
      y_ent,            # y salida o target
      batch_size = nrow(x_ent), # para descenso en gradiente
      callbacks = callback_tensorboard(name),
      view_metrics = FALSE,
      epochs = 20 # número de iteraciones
      )
}

Epoch 1/20

1/1 [==============================] - 0s 26us/step - loss: 1.7096 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 4ms/step - loss: 1.7096 - accuracy: 0.0000e+00 
Epoch 2/20

1/1 [==============================] - 0s 13us/step - loss: 0.1943 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1943 - accuracy: 0.0000e+00 
Epoch 3/20

1/1 [==============================] - 0s 12us/step - loss: 0.1473 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1473 - accuracy: 0.0000e+00 
Epoch 4/20

1/1 [==============================] - 0s 14us/step - loss: 0.1179 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1179 - accuracy: 0.0000e+00 
Epoch 5/20

1/1 [==============================] - 0s 11us/step - loss: 0.0976 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0976 - accuracy: 0.0000e+00 
Epoch 6/20

1/1 [==============================] - 0s 11us/step - loss: 0.0832 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0832 - accuracy: 0.0000e+00 
Epoch 7/20

1/1 [==============================] - 0s 11us/step - loss: 0.0729 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0729 - accuracy: 0.0000e+00 
Epoch 8/20

1/1 [==============================] - 0s 14us/step - loss: 0.0655 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0655 - accuracy: 0.0000e+00 
Epoch 9/20

1/1 [==============================] - 0s 10us/step - loss: 0.0602 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0602 - accuracy: 0.0000e+00 
Epoch 10/20

1/1 [==============================] - 0s 10us/step - loss: 0.0564 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0564 - accuracy: 0.0000e+00 
Epoch 11/20

1/1 [==============================] - 0s 11us/step - loss: 0.0537 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0537 - accuracy: 0.0000e+00 
Epoch 12/20

1/1 [==============================] - 0s 10us/step - loss: 0.0517 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0517 - accuracy: 0.0000e+00 
Epoch 13/20

1/1 [==============================] - 0s 10us/step - loss: 0.0503 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0503 - accuracy: 0.0000e+00 
Epoch 14/20

1/1 [==============================] - 0s 11us/step - loss: 0.0493 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0493 - accuracy: 0.0000e+00 
Epoch 15/20

1/1 [==============================] - 0s 11us/step - loss: 0.0486 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0486 - accuracy: 0.0000e+00 
Epoch 16/20

1/1 [==============================] - 0s 10us/step - loss: 0.0481 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0481 - accuracy: 0.0000e+00 
Epoch 17/20

1/1 [==============================] - 0s 10us/step - loss: 0.0477 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0477 - accuracy: 0.0000e+00 
Epoch 18/20

1/1 [==============================] - 0s 10us/step - loss: 0.0474 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0474 - accuracy: 0.0000e+00 
Epoch 19/20

1/1 [==============================] - 0s 10us/step - loss: 0.0472 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0472 - accuracy: 0.0000e+00 
Epoch 20/20

1/1 [==============================] - 0s 10us/step - loss: 0.0471 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0471 - accuracy: 0.0000e+00 
Epoch 1/20

1/1 [==============================] - 0s 17us/step - loss: 1.7096 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 1.7096 - accuracy: 0.0000e+00 
Epoch 2/20

1/1 [==============================] - 0s 11us/step - loss: 0.1943 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1943 - accuracy: 0.0000e+00 
Epoch 3/20

1/1 [==============================] - 0s 11us/step - loss: 0.1473 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.1473 - accuracy: 0.0000e+00 
Epoch 4/20

1/1 [==============================] - 0s 10us/step - loss: 0.1179 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.1179 - accuracy: 0.0000e+00 
Epoch 5/20

1/1 [==============================] - 0s 10us/step - loss: 0.0976 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0976 - accuracy: 0.0000e+00 
Epoch 6/20

1/1 [==============================] - 0s 10us/step - loss: 0.0832 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0832 - accuracy: 0.0000e+00 
Epoch 7/20

1/1 [==============================] - 0s 10us/step - loss: 0.0729 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0729 - accuracy: 0.0000e+00 
Epoch 8/20

1/1 [==============================] - 0s 10us/step - loss: 0.0655 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0655 - accuracy: 0.0000e+00 
Epoch 9/20

1/1 [==============================] - 0s 10us/step - loss: 0.0602 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0602 - accuracy: 0.0000e+00 
Epoch 10/20

1/1 [==============================] - 0s 11us/step - loss: 0.0564 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0564 - accuracy: 0.0000e+00 
Epoch 11/20

1/1 [==============================] - 0s 10us/step - loss: 0.0537 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0537 - accuracy: 0.0000e+00 
Epoch 12/20

1/1 [==============================] - 0s 11us/step - loss: 0.0517 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0517 - accuracy: 0.0000e+00 
Epoch 13/20

1/1 [==============================] - 0s 10us/step - loss: 0.0503 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0503 - accuracy: 0.0000e+00 
Epoch 14/20

1/1 [==============================] - 0s 11us/step - loss: 0.0493 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0493 - accuracy: 0.0000e+00 
Epoch 15/20

1/1 [==============================] - 0s 14us/step - loss: 0.0486 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0486 - accuracy: 0.0000e+00 
Epoch 16/20

1/1 [==============================] - 0s 11us/step - loss: 0.0481 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0481 - accuracy: 0.0000e+00 
Epoch 17/20

1/1 [==============================] - 0s 10us/step - loss: 0.0477 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0477 - accuracy: 0.0000e+00 
Epoch 18/20

1/1 [==============================] - 0s 10us/step - loss: 0.0474 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0474 - accuracy: 0.0000e+00 
Epoch 19/20

1/1 [==============================] - 0s 10us/step - loss: 0.0472 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0472 - accuracy: 0.0000e+00 
Epoch 20/20

1/1 [==============================] - 0s 10us/step - loss: 0.0471 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0471 - accuracy: 0.0000e+00 
Epoch 1/20

1/1 [==============================] - 0s 17us/step - loss: 1.7096 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 1.7096 - accuracy: 0.0000e+00 
Epoch 2/20

1/1 [==============================] - 0s 14us/step - loss: 0.1943 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1943 - accuracy: 0.0000e+00 
Epoch 3/20

1/1 [==============================] - 0s 13us/step - loss: 0.1473 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1473 - accuracy: 0.0000e+00 
Epoch 4/20

1/1 [==============================] - 0s 11us/step - loss: 0.1179 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.1179 - accuracy: 0.0000e+00 
Epoch 5/20

1/1 [==============================] - 0s 10us/step - loss: 0.0976 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0976 - accuracy: 0.0000e+00 
Epoch 6/20

1/1 [==============================] - 0s 11us/step - loss: 0.0832 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0832 - accuracy: 0.0000e+00 
Epoch 7/20

1/1 [==============================] - 0s 10us/step - loss: 0.0729 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 15ms/step - loss: 0.0729 - accuracy: 0.0000e+00
Epoch 8/20

1/1 [==============================] - 0s 15us/step - loss: 0.0655 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0655 - accuracy: 0.0000e+00 
Epoch 9/20

1/1 [==============================] - 0s 23us/step - loss: 0.0602 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 4ms/step - loss: 0.0602 - accuracy: 0.0000e+00 
Epoch 10/20

1/1 [==============================] - 0s 12us/step - loss: 0.0564 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0564 - accuracy: 0.0000e+00 
Epoch 11/20

1/1 [==============================] - 0s 11us/step - loss: 0.0537 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0537 - accuracy: 0.0000e+00 
Epoch 12/20

1/1 [==============================] - 0s 10us/step - loss: 0.0517 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0517 - accuracy: 0.0000e+00 
Epoch 13/20

1/1 [==============================] - 0s 10us/step - loss: 0.0503 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0503 - accuracy: 0.0000e+00 
Epoch 14/20

1/1 [==============================] - 0s 9us/step - loss: 0.0493 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0493 - accuracy: 0.0000e+00
Epoch 15/20

1/1 [==============================] - 0s 9us/step - loss: 0.0486 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0486 - accuracy: 0.0000e+00
Epoch 16/20

1/1 [==============================] - 0s 10us/step - loss: 0.0481 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0481 - accuracy: 0.0000e+00 
Epoch 17/20

1/1 [==============================] - 0s 10us/step - loss: 0.0477 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0477 - accuracy: 0.0000e+00 
Epoch 18/20

1/1 [==============================] - 0s 9us/step - loss: 0.0474 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0474 - accuracy: 0.0000e+00
Epoch 19/20

1/1 [==============================] - 0s 9us/step - loss: 0.0472 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0472 - accuracy: 0.0000e+00
Epoch 20/20

1/1 [==============================] - 0s 9us/step - loss: 0.0471 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0471 - accuracy: 0.0000e+00
Epoch 1/20

1/1 [==============================] - 0s 22us/step - loss: 1.7096 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 1.7096 - accuracy: 0.0000e+00 
Epoch 2/20

1/1 [==============================] - 0s 13us/step - loss: 0.1943 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1943 - accuracy: 0.0000e+00 
Epoch 3/20

1/1 [==============================] - 0s 12us/step - loss: 0.1473 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1473 - accuracy: 0.0000e+00 
Epoch 4/20

1/1 [==============================] - 0s 10us/step - loss: 0.1179 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1179 - accuracy: 0.0000e+00 
Epoch 5/20

1/1 [==============================] - 0s 11us/step - loss: 0.0976 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0976 - accuracy: 0.0000e+00 
Epoch 6/20

1/1 [==============================] - 0s 10us/step - loss: 0.0832 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0832 - accuracy: 0.0000e+00 
Epoch 7/20

1/1 [==============================] - 0s 10us/step - loss: 0.0729 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0729 - accuracy: 0.0000e+00 
Epoch 8/20

1/1 [==============================] - 0s 10us/step - loss: 0.0655 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0655 - accuracy: 0.0000e+00 
Epoch 9/20

1/1 [==============================] - 0s 10us/step - loss: 0.0602 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0602 - accuracy: 0.0000e+00 
Epoch 10/20

1/1 [==============================] - 0s 10us/step - loss: 0.0564 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0564 - accuracy: 0.0000e+00 
Epoch 11/20

1/1 [==============================] - 0s 9us/step - loss: 0.0537 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0537 - accuracy: 0.0000e+00
Epoch 12/20

1/1 [==============================] - 0s 10us/step - loss: 0.0517 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0517 - accuracy: 0.0000e+00 
Epoch 13/20

1/1 [==============================] - 0s 10us/step - loss: 0.0503 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0503 - accuracy: 0.0000e+00 
Epoch 14/20

1/1 [==============================] - 0s 10us/step - loss: 0.0493 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0493 - accuracy: 0.0000e+00 
Epoch 15/20

1/1 [==============================] - 0s 10us/step - loss: 0.0486 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0486 - accuracy: 0.0000e+00 
Epoch 16/20

1/1 [==============================] - 0s 10us/step - loss: 0.0481 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0481 - accuracy: 0.0000e+00 
Epoch 17/20

1/1 [==============================] - 0s 10us/step - loss: 0.0477 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0477 - accuracy: 0.0000e+00 
Epoch 18/20

1/1 [==============================] - 0s 10us/step - loss: 0.0474 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0474 - accuracy: 0.0000e+00 
Epoch 19/20

1/1 [==============================] - 0s 10us/step - loss: 0.0472 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0472 - accuracy: 0.0000e+00 
Epoch 20/20

1/1 [==============================] - 0s 10us/step - loss: 0.0471 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 23ms/step - loss: 0.0471 - accuracy: 0.0000e+00
Epoch 1/20

1/1 [==============================] - 0s 17us/step - loss: 1.7096 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 1.7096 - accuracy: 0.0000e+00 
Epoch 2/20

1/1 [==============================] - 0s 13us/step - loss: 0.1943 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1943 - accuracy: 0.0000e+00 
Epoch 3/20

1/1 [==============================] - 0s 11us/step - loss: 0.1473 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1473 - accuracy: 0.0000e+00 
Epoch 4/20

1/1 [==============================] - 0s 12us/step - loss: 0.1179 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.1179 - accuracy: 0.0000e+00 
Epoch 5/20

1/1 [==============================] - 0s 10us/step - loss: 0.0976 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0976 - accuracy: 0.0000e+00 
Epoch 6/20

1/1 [==============================] - 0s 10us/step - loss: 0.0832 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0832 - accuracy: 0.0000e+00 
Epoch 7/20

1/1 [==============================] - 0s 10us/step - loss: 0.0729 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0729 - accuracy: 0.0000e+00 
Epoch 8/20

1/1 [==============================] - 0s 10us/step - loss: 0.0655 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0655 - accuracy: 0.0000e+00 
Epoch 9/20

1/1 [==============================] - 0s 10us/step - loss: 0.0602 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0602 - accuracy: 0.0000e+00 
Epoch 10/20

1/1 [==============================] - 0s 10us/step - loss: 0.0564 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0564 - accuracy: 0.0000e+00 
Epoch 11/20

1/1 [==============================] - 0s 9us/step - loss: 0.0537 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0537 - accuracy: 0.0000e+00
Epoch 12/20

1/1 [==============================] - 0s 10us/step - loss: 0.0517 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0517 - accuracy: 0.0000e+00 
Epoch 13/20

1/1 [==============================] - 0s 11us/step - loss: 0.0503 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0503 - accuracy: 0.0000e+00 
Epoch 14/20

1/1 [==============================] - 0s 17us/step - loss: 0.0493 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0493 - accuracy: 0.0000e+00 
Epoch 15/20

1/1 [==============================] - 0s 14us/step - loss: 0.0486 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0486 - accuracy: 0.0000e+00 
Epoch 16/20

1/1 [==============================] - 0s 12us/step - loss: 0.0481 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0481 - accuracy: 0.0000e+00 
Epoch 17/20

1/1 [==============================] - 0s 11us/step - loss: 0.0477 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0477 - accuracy: 0.0000e+00 
Epoch 18/20

1/1 [==============================] - 0s 10us/step - loss: 0.0474 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 2ms/step - loss: 0.0474 - accuracy: 0.0000e+00 
Epoch 19/20

1/1 [==============================] - 0s 11us/step - loss: 0.0472 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0472 - accuracy: 0.0000e+00 
Epoch 20/20

1/1 [==============================] - 0s 10us/step - loss: 0.0471 - accuracy: 0.0000e+00

1/1 [==============================] - 0s 3ms/step - loss: 0.0471 - accuracy: 0.0000e+00 

TensorBoard 2.2.2 at http://127.0.0.1:5001/ (Press CTRL+C to quit)

Started TensorBoard at http://127.0.0.1:5001