With the abundance of nice libraries, in R, for statistical computing, why would you be excited by TensorFlow Likelihood (TFP, for brief)? Nicely – let’s have a look at an inventory of its elements:

• Distributions and bijectors (bijectors are reversible, composable maps)
• Probabilistic modeling (Edward2 and probabilistic community layers)
• Probabilistic inference (by way of MCMC or variational inference)

Now think about all these working seamlessly with the TensorFlow framework – core, Keras, contributed modules – and in addition, working distributed and on GPU. The sphere of potential purposes is huge – and much too numerous to cowl as a complete in an introductory weblog publish.

As a substitute, our intention right here is to offer a primary introduction to TFP, specializing in direct applicability to and interoperability with deep studying.
We’ll shortly present find out how to get began with one of many fundamental constructing blocks: distributions. Then, we’ll construct a variational autoencoder just like that in Illustration studying with MMD-VAE. This time although, we’ll make use of TFP to pattern from the prior and approximate posterior distributions.

We’ll regard this publish as a “proof on idea” for utilizing TFP with Keras – from R – and plan to comply with up with extra elaborate examples from the realm of semi-supervised illustration studying.

To put in TFP along with TensorFlow, merely append tensorflow-probability to the default record of additional packages:

library(tensorflow)
install_tensorflow(
  extra_packages = c("keras", "tensorflow-hub", "tensorflow-probability"),
  model = "1.12"
)

Now to make use of TFP, all we have to do is import it and create some helpful handles.

And right here we go, sampling from a normal regular distribution.

n <- tfd$Regular(loc = 0, scale = 1)
n$pattern(6L)

tf.Tensor(
"Normal_1/pattern/Reshape:0", form=(6,), dtype=float32
)

Now that’s good, however it’s 2019, we don’t wish to need to create a session to judge these tensors anymore. Within the variational autoencoder instance beneath, we’re going to see how TFP and TF keen execution are the right match, so why not begin utilizing it now.

To make use of keen execution, we now have to execute the next traces in a recent (R) session:

… and import TFP, similar as above.

tfp <- import("tensorflow_probability")
tfd <- tfp$distributions

Now let’s shortly have a look at TFP distributions.

Utilizing distributions

Right here’s that normal regular once more.

n <- tfd$Regular(loc = 0, scale = 1)

Issues generally finished with a distribution embrace sampling:

# simply as in low-level tensorflow, we have to append L to point integer arguments
n$pattern(6L) 

tf.Tensor(
[-0.34403768 -0.14122334 -1.3832929   1.618252    1.364448   -1.1299014 ],
form=(6,),
dtype=float32
)

In addition to getting the log chance. Right here we do this concurrently for 3 values.

tf.Tensor(
[-1.4189385 -0.9189385 -1.4189385], form=(3,), dtype=float32
)

We will do the identical issues with numerous different distributions, e.g., the Bernoulli:

b <- tfd$Bernoulli(0.9)
b$pattern(10L)

tf.Tensor(
[1 1 1 0 1 1 0 1 0 1], form=(10,), dtype=int32
)

tf.Tensor(
[-1.2411538 -0.3411539 -1.2411538 -1.2411538], form=(4,), dtype=float32
)

Observe that within the final chunk, we’re asking for the log chances of 4 unbiased attracts.

Batch shapes and occasion shapes

In TFP, we are able to do the next.

ns <- tfd$Regular(
  loc = c(1, 10, -200),
  scale = c(0.1, 0.1, 1)
)
ns

tfp.distributions.Regular(
"Regular/", batch_shape=(3,), event_shape=(), dtype=float32
)

Opposite to what it would appear like, this isn’t a multivariate regular. As indicated by batch_shape=(3,), it is a “batch” of unbiased univariate distributions. The truth that these are univariate is seen in event_shape=(): Every of them lives in one-dimensional occasion area.

If as an alternative we create a single, two-dimensional multivariate regular:

n <- tfd$MultivariateNormalDiag(loc = c(0, 10), scale_diag = c(1, 4))
n

tfp.distributions.MultivariateNormalDiag(
"MultivariateNormalDiag/", batch_shape=(), event_shape=(2,), dtype=float32
)

we see batch_shape=(), event_shape=(2,), as anticipated.

In fact, we are able to mix each, creating batches of multivariate distributions:

nd_batch <- tfd$MultivariateNormalFullCovariance(
  loc = record(c(0., 0.), c(1., 1.), c(2., 2.)),
  covariance_matrix = record(
    matrix(c(1, .1, .1, 1), ncol = 2),
    matrix(c(1, .3, .3, 1), ncol = 2),
    matrix(c(1, .5, .5, 1), ncol = 2))
)

This instance defines a batch of three two-dimensional multivariate regular distributions.

Changing between batch shapes and occasion shapes

Unusual as it might sound, conditions come up the place we wish to rework distribution shapes between these varieties – in reality, we’ll see such a case very quickly.

tfd$Impartial is used to transform dimensions in batch_shape to dimensions in event_shape.

Here’s a batch of three unbiased Bernoulli distributions.

bs <- tfd$Bernoulli(probs=c(.3,.5,.7))
bs

tfp.distributions.Bernoulli(
"Bernoulli/", batch_shape=(3,), event_shape=(), dtype=int32
)

We will convert this to a digital “three-dimensional” Bernoulli like this:

b <- tfd$Impartial(bs, reinterpreted_batch_ndims = 1L)
b

tfp.distributions.Impartial(
"IndependentBernoulli/", batch_shape=(), event_shape=(3,), dtype=int32
)

Right here reinterpreted_batch_ndims tells TFP how lots of the batch dimensions are getting used for the occasion area, beginning to rely from the fitting of the form record.

With this fundamental understanding of TFP distributions, we’re able to see them utilized in a VAE.

We’ll take the (not so) deep convolutional structure from Illustration studying with MMD-VAE and use distributions for sampling and computing chances. Optionally, our new VAE will be capable to be taught the prior distribution.

Concretely, the next exposition will include three elements.
First, we current widespread code relevant to each a VAE with a static prior, and one which learns the parameters of the prior distribution.
Then, we now have the coaching loop for the primary (static-prior) VAE. Lastly, we focus on the coaching loop and extra mannequin concerned within the second (prior-learning) VAE.

Presenting each variations one after the opposite results in code duplications, however avoids scattering complicated if-else branches all through the code.

The second VAE is offered as a part of the Keras examples so that you don’t have to repeat out code snippets. The code additionally incorporates further performance not mentioned and replicated right here, reminiscent of for saving mannequin weights.

So, let’s begin with the widespread half.

On the threat of repeating ourselves, right here once more are the preparatory steps (together with just a few further library masses).

Dataset

For a change from MNIST and Vogue-MNIST, we’ll use the model new Kuzushiji-MNIST(Clanuwat et al. 2018).

np <- import("numpy")

kuzushiji <- np$load("kmnist-train-imgs.npz")
kuzushiji <- kuzushiji$get("arr_0")
 
train_images <- kuzushiji %>%
  k_expand_dims() %>%
  k_cast(dtype = "float32")

train_images <- train_images %>% `/`(255)

As in that different publish, we stream the info by way of tfdatasets:

buffer_size <- 60000
batch_size <- 256
batches_per_epoch <- buffer_size / batch_size

train_dataset <- tensor_slices_dataset(train_images) %>%
  dataset_shuffle(buffer_size) %>%
  dataset_batch(batch_size)

Now let’s see what adjustments within the encoder and decoder fashions.

Encoder

The encoder differs from what we had with out TFP in that it doesn’t return the approximate posterior means and variances immediately as tensors. As a substitute, it returns a batch of multivariate regular distributions:

# you would possibly wish to change this relying on the dataset
latent_dim <- 2

encoder_model <- operate(identify = NULL) {

  keras_model_custom(identify = identify, operate(self) {
  
    self$conv1 <-
      layer_conv_2d(
        filters = 32,
        kernel_size = 3,
        strides = 2,
        activation = "relu"
      )
    self$conv2 <-
      layer_conv_2d(
        filters = 64,
        kernel_size = 3,
        strides = 2,
        activation = "relu"
      )
    self$flatten <- layer_flatten()
    self$dense <- layer_dense(models = 2 * latent_dim)
    
    operate (x, masks = NULL) {
      x <- x %>%
        self$conv1() %>%
        self$conv2() %>%
        self$flatten() %>%
        self$dense()
        
      tfd$MultivariateNormalDiag(
        loc = x[, 1:latent_dim],
        scale_diag = tf$nn$softplus(x[, (latent_dim + 1):(2 * latent_dim)] + 1e-5)
      )
    }
  })
}

Let’s do this out.

encoder <- encoder_model()

iter <- make_iterator_one_shot(train_dataset)
x <-  iterator_get_next(iter)

approx_posterior <- encoder(x)
approx_posterior

tfp.distributions.MultivariateNormalDiag(
"MultivariateNormalDiag/", batch_shape=(256,), event_shape=(2,), dtype=float32
)

approx_posterior$pattern()

tf.Tensor(
[[ 5.77791929e-01 -1.64988488e-02]
 [ 7.93901443e-01 -1.00042784e+00]
 [-1.56279251e-01 -4.06365871e-01]
 ...
 ...
 [-6.47531569e-01  2.10889503e-02]], form=(256, 2), dtype=float32)

We don’t find out about you, however we nonetheless benefit from the ease of inspecting values with keen execution – loads.

Now, on to the decoder, which too returns a distribution as an alternative of a tensor.

Decoder

Within the decoder, we see why transformations between batch form and occasion form are helpful.
The output of self$deconv3 is four-dimensional. What we want is an on-off-probability for each pixel.
Previously, this was completed by feeding the tensor right into a dense layer and making use of a sigmoid activation.
Right here, we use tfd$Impartial to successfully tranform the tensor right into a chance distribution over three-dimensional photographs (width, top, channel(s)).

decoder_model <- operate(identify = NULL) {
  
  keras_model_custom(identify = identify, operate(self) {
    
    self$dense <- layer_dense(models = 7 * 7 * 32, activation = "relu")
    self$reshape <- layer_reshape(target_shape = c(7, 7, 32))
    self$deconv1 <-
      layer_conv_2d_transpose(
        filters = 64,
        kernel_size = 3,
        strides = 2,
        padding = "similar",
        activation = "relu"
      )
    self$deconv2 <-
      layer_conv_2d_transpose(
        filters = 32,
        kernel_size = 3,
        strides = 2,
        padding = "similar",
        activation = "relu"
      )
    self$deconv3 <-
      layer_conv_2d_transpose(
        filters = 1,
        kernel_size = 3,
        strides = 1,
        padding = "similar"
      )
    
    operate (x, masks = NULL) {
      x <- x %>%
        self$dense() %>%
        self$reshape() %>%
        self$deconv1() %>%
        self$deconv2() %>%
        self$deconv3()
      
      tfd$Impartial(tfd$Bernoulli(logits = x),
                      reinterpreted_batch_ndims = 3L)
      
    }
  })
}

Let’s do this out too.

decoder <- decoder_model()
decoder_likelihood <- decoder(approx_posterior_sample)

tfp.distributions.Impartial(
"IndependentBernoulli/", batch_shape=(256,), event_shape=(28, 28, 1), dtype=int32
)

This distribution will probably be used to generate the “reconstructions,” in addition to decide the loglikelihood of the unique samples.

KL loss and optimizer

Each VAEs mentioned beneath will want an optimizer …

optimizer <- tf$practice$AdamOptimizer(1e-4)

… and each will delegate to compute_kl_loss to compute the KL a part of the loss.

This helper operate merely subtracts the log probability of the samples beneath the prior from their loglikelihood beneath the approximate posterior.

compute_kl_loss <- operate(
  latent_prior,
  approx_posterior,
  approx_posterior_sample) {
  
  kl_div <- approx_posterior$log_prob(approx_posterior_sample) -
    latent_prior$log_prob(approx_posterior_sample)
  avg_kl_div <- tf$reduce_mean(kl_div)
  avg_kl_div
}

Now that we’ve appeared on the widespread elements, we first focus on find out how to practice a VAE with a static prior.

On this VAE, we use TFP to create the standard isotropic Gaussian prior.
We then immediately pattern from this distribution within the coaching loop.

latent_prior <- tfd$MultivariateNormalDiag(
  loc  = tf$zeros(record(latent_dim)),
  scale_identity_multiplier = 1
)

And right here is the whole coaching loop. We’ll level out the essential TFP-related steps beneath.

for (epoch in seq_len(num_epochs)) {
  iter <- make_iterator_one_shot(train_dataset)
  
  total_loss <- 0
  total_loss_nll <- 0
  total_loss_kl <- 0
  
  until_out_of_range({
    x <-  iterator_get_next(iter)
    
    with(tf$GradientTape(persistent = TRUE) %as% tape, {
      approx_posterior <- encoder(x)
      approx_posterior_sample <- approx_posterior$pattern()
      decoder_likelihood <- decoder(approx_posterior_sample)
      
      nll <- -decoder_likelihood$log_prob(x)
      avg_nll <- tf$reduce_mean(nll)
      
      kl_loss <- compute_kl_loss(
        latent_prior,
        approx_posterior,
        approx_posterior_sample
      )

      loss <- kl_loss + avg_nll
    })
    
    total_loss <- total_loss + loss
    total_loss_nll <- total_loss_nll + avg_nll
    total_loss_kl <- total_loss_kl + kl_loss
    
    encoder_gradients <- tape$gradient(loss, encoder$variables)
    decoder_gradients <- tape$gradient(loss, decoder$variables)
    
    optimizer$apply_gradients(purrr::transpose(record(
      encoder_gradients, encoder$variables
    )),
    global_step = tf$practice$get_or_create_global_step())
    optimizer$apply_gradients(purrr::transpose(record(
      decoder_gradients, decoder$variables
    )),
    global_step = tf$practice$get_or_create_global_step())
 
  })
  
  cat(
    glue(
      "Losses (epoch): {epoch}:",
      "  {(as.numeric(total_loss_nll)/batches_per_epoch) %>% spherical(4)} nll",
      "  {(as.numeric(total_loss_kl)/batches_per_epoch) %>% spherical(4)} kl",
      "  {(as.numeric(total_loss)/batches_per_epoch) %>% spherical(4)} complete"
    ),
    "n"
  )
}

Above, enjoying round with the encoder and the decoder, we’ve already seen how

approx_posterior <- encoder(x)

provides us a distribution we are able to pattern from. We use it to acquire samples from the approximate posterior:

approx_posterior_sample <- approx_posterior$pattern()

These samples, we take them and feed them to the decoder, who provides us on-off-likelihoods for picture pixels.

decoder_likelihood <- decoder(approx_posterior_sample)

Now the loss consists of the standard ELBO elements: reconstruction loss and KL divergence.
The reconstruction loss we immediately get hold of from TFP, utilizing the realized decoder distribution to evaluate the probability of the unique enter.

nll <- -decoder_likelihood$log_prob(x)
avg_nll <- tf$reduce_mean(nll)

The KL loss we get from compute_kl_loss, the helper operate we noticed above:

kl_loss <- compute_kl_loss(
        latent_prior,
        approx_posterior,
        approx_posterior_sample
      )

We add each and arrive on the total VAE loss:

loss <- kl_loss + avg_nll

Aside from these adjustments on account of utilizing TFP, the coaching course of is simply regular backprop, the best way it seems utilizing keen execution.

Now let’s see how as an alternative of utilizing the usual isotropic Gaussian, we might be taught a mix of Gaussians.
The selection of variety of distributions right here is fairly arbitrary. Simply as with latent_dim, you would possibly wish to experiment and discover out what works greatest in your dataset.

mixture_components <- 16

learnable_prior_model <- operate(identify = NULL, latent_dim, mixture_components) {
  
  keras_model_custom(identify = identify, operate(self) {
    
    self$loc <-
      tf$get_variable(
        identify = "loc",
        form = record(mixture_components, latent_dim),
        dtype = tf$float32
      )
    self$raw_scale_diag <- tf$get_variable(
      identify = "raw_scale_diag",
      form = c(mixture_components, latent_dim),
      dtype = tf$float32
    )
    self$mixture_logits <-
      tf$get_variable(
        identify = "mixture_logits",
        form = c(mixture_components),
        dtype = tf$float32
      )
      
    operate (x, masks = NULL) {
        tfd$MixtureSameFamily(
          components_distribution = tfd$MultivariateNormalDiag(
            loc = self$loc,
            scale_diag = tf$nn$softplus(self$raw_scale_diag)
          ),
          mixture_distribution = tfd$Categorical(logits = self$mixture_logits)
        )
      }
    })
  }

In TFP terminology, components_distribution is the underlying distribution sort, and mixture_distribution holds the chances that particular person elements are chosen.

Observe how self$loc, self$raw_scale_diag and self$mixture_logits are TensorFlow Variables and thus, persistent and updatable by backprop.

Now we create the mannequin.

latent_prior_model <- learnable_prior_model(
  latent_dim = latent_dim,
  mixture_components = mixture_components
)

How can we get hold of a latent prior distribution we are able to pattern from? A bit unusually, this mannequin will probably be referred to as with out an enter:

latent_prior <- latent_prior_model(NULL)
latent_prior

tfp.distributions.MixtureSameFamily(
"MixtureSameFamily/", batch_shape=(), event_shape=(2,), dtype=float32
)

Right here now could be the whole coaching loop. Observe how we now have a 3rd mannequin to backprop via.

for (epoch in seq_len(num_epochs)) {
  iter <- make_iterator_one_shot(train_dataset)
  
  total_loss <- 0
  total_loss_nll <- 0
  total_loss_kl <- 0
  
  until_out_of_range({
    x <-  iterator_get_next(iter)
    
    with(tf$GradientTape(persistent = TRUE) %as% tape, {
      approx_posterior <- encoder(x)
      
      approx_posterior_sample <- approx_posterior$pattern()
      decoder_likelihood <- decoder(approx_posterior_sample)
      
      nll <- -decoder_likelihood$log_prob(x)
      avg_nll <- tf$reduce_mean(nll)
      
      latent_prior <- latent_prior_model(NULL)
      
      kl_loss <- compute_kl_loss(
        latent_prior,
        approx_posterior,
        approx_posterior_sample
      )

      loss <- kl_loss + avg_nll
    })
    
    total_loss <- total_loss + loss
    total_loss_nll <- total_loss_nll + avg_nll
    total_loss_kl <- total_loss_kl + kl_loss
    
    encoder_gradients <- tape$gradient(loss, encoder$variables)
    decoder_gradients <- tape$gradient(loss, decoder$variables)
    prior_gradients <-
      tape$gradient(loss, latent_prior_model$variables)
    
    optimizer$apply_gradients(purrr::transpose(record(
      encoder_gradients, encoder$variables
    )),
    global_step = tf$practice$get_or_create_global_step())
    optimizer$apply_gradients(purrr::transpose(record(
      decoder_gradients, decoder$variables
    )),
    global_step = tf$practice$get_or_create_global_step())
    optimizer$apply_gradients(purrr::transpose(record(
      prior_gradients, latent_prior_model$variables
    )),
    global_step = tf$practice$get_or_create_global_step())
    
  })
  
  checkpoint$save(file_prefix = checkpoint_prefix)
  
  cat(
    glue(
      "Losses (epoch): {epoch}:",
      "  {(as.numeric(total_loss_nll)/batches_per_epoch) %>% spherical(4)} nll",
      "  {(as.numeric(total_loss_kl)/batches_per_epoch) %>% spherical(4)} kl",
      "  {(as.numeric(total_loss)/batches_per_epoch) %>% spherical(4)} complete"
    ),
    "n"
  )
}  

And that’s it! For us, each VAEs yielded comparable outcomes, and we didn’t expertise nice variations from experimenting with latent dimensionality and the variety of combination distributions. However once more, we wouldn’t wish to generalize to different datasets, architectures, and so forth.

Talking of outcomes, how do they appear? Right here we see letters generated after 40 epochs of coaching. On the left are random letters, on the fitting, the standard VAE grid show of latent area.

Hopefully, we’ve succeeded in exhibiting that TensorFlow Likelihood, keen execution, and Keras make for a lovely mixture! In case you relate complete quantity of code required to the complexity of the duty, in addition to depth of the ideas concerned, this could seem as a fairly concise implementation.

Within the nearer future, we plan to comply with up with extra concerned purposes of TensorFlow Likelihood, principally from the realm of illustration studying. Keep tuned!