rstudio / ai-blog Goto Github PK
View Code? Open in Web Editor NEWRepository for the RStudio AI Blog (formerly: TensorFlow for R Blog)
Home Page: https://blogs.rstudio.com/ai/
Repository for the RStudio AI Blog (formerly: TensorFlow for R Blog)
Home Page: https://blogs.rstudio.com/ai/
We are proposing a post entitled The text-package for the RStudio AI blog.
We have followed the instructions from https://blogs.rstudio.com/ai/contributing.html.
Our proposed blog post can be found at https://github.com/OscarKjell/ai-blog/tree/text_huggingface_in_r/_posts/2022-09-29-r-text
Looking forward to your feedback.
Kind Regards,
Oscar
I don't see a button/link I could click to subscribe (on Firefox)
<div id="subscribe-caption" style="line-height: 1.2; margin-bottom: 2px; width: 377px; font-size: 15px;">
Enjoy this blog? Get notified of new posts by email:
</div>
Could you add URL to RStudio AI Blog in this repository in About section?
Hi, I'm having trouble in installing a good version of tensorflow on my laptop.
First I installed a basic version of it (just using install_tensorflow()) but when I want to confirm that the installation succeeded:
_> library(tensorflow)
sess = tf$Session()
2018-08-28 12:06:53.362284: I T:\src\github\tensorflow\tensorflow\core\platform\cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
hello <- tf$constant('Hello, TensorFlow!')
sess$run(hello)
b'Hello, TensorFlow!'_
So I ry removing myconda environment and then re-installing with:
_library(reticulate)
conda_remove("r-tensorflow")
install.packages("tensorflow")
library(tensorflow)
install_tensorflow(version = "somekindofversion")_
but I get this error: Error: Error 2 occurred installing packages into conda environment r-tensorflow
This is the entire script:
install_tensorflow(version = "https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.4.0/py36/CPU/avx2/tensorflow-1.4.0-cp36-cp36m-win_amd64.whl")
Creating r-tensorflow conda environment for TensorFlow installation...
Solving environment: ...working... done
environment location: G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow
added / updated specs:
- python=3.6
The following NEW packages will be INSTALLED:
certifi: 2018.8.24-py36_1
pip: 10.0.1-py36_0
python: 3.6.6-hea74fb7_0
setuptools: 40.2.0-py36_0
vc: 14.1-h0510ff6_3
vs2015_runtime: 15.5.2-3
wheel: 0.31.1-py36_0
wincertstore: 0.2-py36h7fe50ca_0
Preparing transaction: ...working... done
Verifying transaction: ...working...
SafetyError: The package for setuptools located at G:\G_PROGRAMMI\Anaconda3\pkgs\setuptools-40.2.0-py36_0
appears to be corrupted. The path 'Scripts/easy_install.exe'
has a sha256 mismatch.
reported sha256: 993203a406e04936a07829b1f482fd27d739b640482e213f4c49ea1ee78a5fcf
actual sha256: 0cc372d47e0e71c25012697ae0a2feca19cfa46caccb872a555c3669e79f701b
SafetyError: The package for wheel located at G:\G_PROGRAMMI\Anaconda3\pkgs\wheel-0.31.1-py36_0
appears to be corrupted. The path 'Scripts/wheel.exe'
has a sha256 mismatch.
reported sha256: 993203a406e04936a07829b1f482fd27d739b640482e213f4c49ea1ee78a5fcf
actual sha256: 2a882e6c7ea316634261d68f7ccb7aadc34fe864b4b834611c9a4cd26efc9d35
done
Executing transaction: ...working... done
Determining latest installable release of TensorFlow...done
Installing TensorFlow...
Collecting tensorflow==1.4.0 from https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.4.0/py36/CPU/avx2/tensorflow-1.4.0-cp36-cp36m-win_amd64.whl
Downloading https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.4.0/py36/CPU/avx2/tensorflow-1.4.0-cp36-cp36m-win_amd64.whl
Exception:
Traceback (most recent call last):
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\basecommand.py", line 228, in main
status = self.run(options, args)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\commands\install.py", line 291, in run
resolver.resolve(requirement_set)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\resolve.py", line 103, in resolve
self._resolve_one(requirement_set, req)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\resolve.py", line 257, in _resolve_one
abstract_dist = self._get_abstract_dist_for(req_to_install)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\resolve.py", line 210, in _get_abstract_dist_for
self.require_hashes
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\operations\prepare.py", line 310, in prepare_linked_requirement
progress_bar=self.progress_bar
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\download.py", line 837, in unpack_url
progress_bar=progress_bar
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\download.py", line 678, in unpack_http_url
unpack_file(from_path, location, content_type, link)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\utils\misc.py", line 575, in unpack_file
flatten=not filename.endswith('.whl')
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\site-packages\pip_internal\utils\misc.py", line 460, in unzip_file
zip = zipfile.ZipFile(zipfp, allowZip64=True)
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\zipfile.py", line 1108, in init
self._RealGetContents()
File "G:\G_PROGRAMMI\Anaconda3\envs\r-tensorflow\lib\zipfile.py", line 1175, in _RealGetContents
raise BadZipFile("File is not a zip file")
zipfile.BadZipFile: File is not a zip file
You are using pip version 10.0.1, however version 18.0 is available.
You should consider upgrading via the 'python -m pip install --upgrade pip' command.
Error: Error 2 occurred installing packages into conda environment r-tensorflow
The version I tried to install is takek from "https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.4.0/py36/CPU/avx2/tensorflow-1.4.0-cp36-cp36m-win_amd64.whl"
Any help would be greatly appreciated, thanks!
Hi,
My lab's server has 2 nvidia GTX 1080 Ti GPUs and when I run the code as shown in this post it takes ~3000 s per epoch with the exact same parameters as in the post. I was wondering if you guys used a different GPU that is much better than the 1080 Ti.
Please let me know if you need me to post the code sample (I literally copy pasted to see how long my Lab's GPU takes to do this task) or anything else!
Thanks
trying to run the code chunks from the blog post:
https://github.com/rstudio/ai-blog/tree/master/_posts/2020-09-29-introducing-torch-for-r
I get reproducible fails at the central code chunk:
for (epoch in 1:5) {
l <- c()
for (b in enumerate(train_dl)) {
# make sure each batch's gradient updates are calculated from a fresh start
optimizer$zero_grad()
# get model predictions
output <- model(b[[1]]$to(device = "cuda"))
# calculate loss
loss <- nnf_cross_entropy(output, b[[2]]$to(device = "cuda"))
# calculate gradient
loss$backward()
# apply weight updates
optimizer$step()
# track losses
l <- c(l, loss$item())
}
cat(sprintf("Loss at epoch %d: %3f\n", epoch, mean(l)))
}
Trying this with packages torch
and torchvision
installed today from github
as suggested at the beginning of the blog.
Fehler in parent.env(x)[["batch"]][[name]] : Objekt des Typs 'symbol' ist nicht indizierbar
40.
`[[.enum_env`(b, 1)
39.
b[[1]]
38.
mget(x = c("input", "weight", "bias", "stride", "padding", "dilation", "groups"))
37.
torch_conv2d(input = input, weight = weight, bias = bias, stride = stride, padding = padding, dilation = dilation, groups = groups)
36.
nnf_conv2d(input, weight, self$bias, self$stride, self$padding, self$dilation, self$groups)
35.
self$conv_forward_(input, self$weight)
34.
self$conv1(.)
33.
mget(x = c("self"))
32.
torch_relu(input)
31.
nnf_relu(.)
30.
mget(x = c("input", "weight", "bias", "stride", "padding", "dilation", "groups"))
29.
torch_conv2d(input = input, weight = weight, bias = bias, stride = stride, padding = padding, dilation = dilation, groups = groups)
28.
nnf_conv2d(input, weight, self$bias, self$stride, self$padding, self$dilation, self$groups)
27.
self$conv_forward_(input, self$weight)
26.
self$conv2(.)
25.
mget(x = c("self"))
24.
torch_relu(input)
23.
nnf_relu(.)
22.
mget(x = c("self", "kernel_size", "stride", "padding", "dilation", "ceil_mode"))
21.
torch_max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)
20.
nnf_max_pool2d(., 2)
19.
mget(x = c("input", "p", "train"))
18.
torch_feature_dropout(input, p, training)
17.
nnf_dropout2d(input, self$p, self$training, self$inplace)
16.
self$dropout1(.)
15.
mget(x = c("self", "dims", "start_dim", "end_dim", "out_dim"))
14.
torch_flatten(., start_dim = 2)
13.
nnf_linear(input, self$weight, self$bias)
12.
self$fc1(.)
11.
mget(x = c("self"))
10.
torch_relu(input)
9.
nnf_relu(.)
8.
mget(x = c("input", "p", "train"))
7.
torch_feature_dropout(input, p, training)
6.
nnf_dropout2d(input, self$p, self$training, self$inplace)
5.
self$dropout2(.)
4.
nnf_linear(input, self$weight, self$bias)
3.
self$fc2(.)
2.
x %>% self$conv1() %>% nnf_relu() %>% self$conv2() %>% nnf_relu() %>% nnf_max_pool2d(2) %>% self$dropout1() %>% torch_flatten(start_dim = 2) %>% self$fc1() %>% nnf_relu() %>% self$dropout2() %>% self$fc2()
1.
model(b[[1]]$to(device = "cuda"))
R version:
R version 4.0.3 (2020-10-10) -- "Bunny-Wunnies Freak Out"
Copyright (C) 2020 The R Foundation for Statistical Computing
Platform: x86_64-suse-linux-gnu (64-bit)
R Studio 1.4.1623 (from the dailies)
rsession using cuda properly has been verified by:
torch::torch_tensor(1, device = "cuda")
and
/usr/local/cuda-10.2/extras/demo_suite> nvidia-smi
Sat Mar 13 14:29:55 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.56 Driver Version: 460.56 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Quadro M1200 Off | 00000000:01:00.0 Off | N/A |
| N/A 49C P0 N/A / N/A | 1054MiB / 4046MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 13898 C .../lib/rstudio/bin/rsession 515MiB |
| 0 N/A N/A 25740 C .../lib/rstudio/bin/rsession 535MiB |
+-----------------------------------------------------------------------------+
I am do not know enough about all this to try to debug that issue by myself. I think this native torch in R package is a great way to get up to date NNs going in R. If my issue with the sample code is a general problem, it might hamper the success of the package, as it is a big hurdle for starting a torch in R journey.
The narative says a
training period of 50 years, a testing period of 10 years, and a skip span of 20 years but the code has a
training period of 100 years, a testing period of 50 years, and skip span of 22 years.
Again, thanks for what you do. 👍
Hello,
I am trying to replicate the results of the following article written by @skeydan :
https://blogs.rstudio.com/ai/posts/2020-07-31-fnn-vae-for-noisy-timeseries/
I would like to get an help on why teh following code :
library(tidyverse)
library(tensorflow)
library(keras)
library(tfdatasets)
library(tfautograph)
library(reticulate)
library(purrr)
library(listarrays)
library(abind)
library(deSolve)
parameters <- c(a = .2,
b = .2,
c = 5.7)
initial_state <-
c(x = 1,
y = 1,
z = 1.05)
roessler <- function(t, state, parameters) {
with(as.list(c(state, parameters)), {
dx <- -y - z
dy <- x + a * y
dz = b + z * (x - c)
list(c(dx, dy, dz))
})
}
times <- seq(0, 2500, length.out = 20000)
roessler_ts <-
ode(
y = initial_state,
times = times,
func = roessler,
parms = parameters,
method = "lsoda"
) %>% unclass() %>% as_tibble()
n <- 10000
roessler <- roessler_ts$x[1:n]
roessler <- scale(roessler)
# add noise
noise <- 1 # also used 1.5, 2, 2.5
roessler <- roessler + rnorm(10000, mean = 0, sd = noise)
str(roessler)
vae_encoder_model <- function(n_timesteps,
n_features,
n_latent,
name = NULL) {
keras_model_custom(name = name, function(self) {
self$conv1 <- layer_conv_1d(kernel_size = 3,
filters = 16,
strides = 2)
self$act1 <- layer_activation_leaky_relu()
self$batchnorm1 <- layer_batch_normalization()
self$conv2 <- layer_conv_1d(kernel_size = 7,
filters = 32,
strides = 2)
self$act2 <- layer_activation_leaky_relu()
self$batchnorm2 <- layer_batch_normalization()
self$conv3 <- layer_conv_1d(kernel_size = 9,
filters = 64,
strides = 2)
self$act3 <- layer_activation_leaky_relu()
self$batchnorm3 <- layer_batch_normalization()
self$conv4 <- layer_conv_1d(
kernel_size = 9,
filters = n_latent,
strides = 2,
activation = "linear"
)
self$batchnorm4 <- layer_batch_normalization()
self$flat <- layer_flatten()
function (x, mask = NULL) {
x %>%
self$conv1() %>%
self$act1() %>%
self$batchnorm1() %>%
self$conv2() %>%
self$act2() %>%
self$batchnorm2() %>%
self$conv3() %>%
self$act3() %>%
self$batchnorm3() %>%
self$conv4() %>%
self$batchnorm4() %>%
self$flat()
}
})
}
vae_decoder_model <- function(n_timesteps,
n_features,
n_latent,
name = NULL) {
keras_model_custom(name = name, function(self) {
self$reshape <- layer_reshape(target_shape = c(1, n_latent))
self$conv1 <- layer_conv_1d_transpose(kernel_size = 15,
filters = 64,
strides = 3)
self$act1 <- layer_activation_leaky_relu()
self$batchnorm1 <- layer_batch_normalization()
self$conv2 <- layer_conv_1d_transpose(kernel_size = 11,
filters = 32,
strides = 3)
self$act2 <- layer_activation_leaky_relu()
self$batchnorm2 <- layer_batch_normalization()
self$conv3 <- layer_conv_1d_transpose(
kernel_size = 9,
filters = 16,
strides = 2,
output_padding = 1
)
self$act3 <- layer_activation_leaky_relu()
self$batchnorm3 <- layer_batch_normalization()
self$conv4 <- layer_conv_1d_transpose(
kernel_size = 7,
filters = 1,
strides = 1,
activation = "linear"
)
self$batchnorm4 <- layer_batch_normalization()
function (x, mask = NULL) {
x %>%
self$reshape() %>%
self$conv1() %>%
self$act1() %>%
self$batchnorm1() %>%
self$conv2() %>%
self$act2() %>%
self$batchnorm2() %>%
self$conv3() %>%
self$act3() %>%
self$batchnorm3() %>%
self$conv4() %>%
self$batchnorm4()
}
})
}
# to reparameterize encoder output before calling decoder
reparameterize <- function(mean, logvar = 0) {
eps <- k_random_normal(shape = n_latent)
eps * k_exp(logvar * 0.5) + mean
}
# loss FNN
loss_false_nn <- function(x) {
# changing these parameters is equivalent to
# changing the strength of the regularizer, so we keep these fixed (these values
# correspond to the original values used in Kennel et al 1992).
rtol <- 10
atol <- 2
k_frac <- 0.01
k <- max(1, floor(k_frac * batch_size))
## Vectorized version of distance matrix calculation
tri_mask <-
tf$linalg$band_part(
tf$ones(
shape = c(tf$cast(n_latent, tf$int32), tf$cast(n_latent, tf$int32)),
dtype = tf$float32
),
num_lower = -1L,
num_upper = 0L
)
# latent x batch_size x latent
batch_masked <-
tf$multiply(tri_mask[, tf$newaxis,], x[tf$newaxis, reticulate::py_ellipsis()])
# latent x batch_size x 1
x_squared <-
tf$reduce_sum(batch_masked * batch_masked,
axis = 2L,
keepdims = TRUE)
# latent x batch_size x batch_size
pdist_vector <- x_squared + tf$transpose(x_squared, perm = c(0L, 2L, 1L)) -
2 * tf$matmul(batch_masked, tf$transpose(batch_masked, perm = c(0L, 2L, 1L)))
#(latent, batch_size, batch_size)
all_dists <- pdist_vector
# latent
all_ra <-
tf$sqrt((1 / (
batch_size * tf$range(1, 1 + n_latent, dtype = tf$float32)
)) *
tf$reduce_sum(tf$square(
batch_masked - tf$reduce_mean(batch_masked, axis = 1L, keepdims = TRUE)
), axis = c(1L, 2L)))
# Avoid singularity in the case of zeros
#(latent, batch_size, batch_size)
all_dists <-
tf$clip_by_value(all_dists, 1e-14, tf$reduce_max(all_dists))
#inds = tf.argsort(all_dists, axis=-1)
top_k <- tf$math$top_k(-all_dists, tf$cast(k + 1, tf$int32))
# (#(latent, batch_size, batch_size)
top_indices <- top_k[[1]]
#(latent, batch_size, batch_size)
neighbor_dists_d <-
tf$gather(all_dists, top_indices, batch_dims = -1L)
#(latent - 1, batch_size, batch_size)
neighbor_new_dists <-
tf$gather(all_dists[2:-1, , ],
top_indices[1:-2, , ],
batch_dims = -1L)
# Eq. 4 of Kennel et al.
#(latent - 1, batch_size, batch_size)
scaled_dist <- tf$sqrt((
tf$square(neighbor_new_dists) -
# (9, 8, 2)
tf$square(neighbor_dists_d[1:-2, , ])) /
# (9, 8, 2)
tf$square(neighbor_dists_d[1:-2, , ])
)
# Kennel condition #1
#(latent - 1, batch_size, batch_size)
is_false_change <- (scaled_dist > rtol)
# Kennel condition 2
#(latent - 1, batch_size, batch_size)
is_large_jump <-
(neighbor_new_dists > atol * all_ra[1:-2, tf$newaxis, tf$newaxis])
is_false_neighbor <-
tf$math$logical_or(is_false_change, is_large_jump)
#(latent - 1, batch_size, 1)
total_false_neighbors <-
tf$cast(is_false_neighbor, tf$int32)[reticulate::py_ellipsis(), 2:(k + 2)]
# Pad zero to match dimensionality of latent space
# (latent - 1)
reg_weights <-
1 - tf$reduce_mean(tf$cast(total_false_neighbors, tf$float32), axis = c(1L, 2L))
# (latent,)
reg_weights <- tf$pad(reg_weights, list(list(1L, 0L)))
# Find batch average activity
# L2 Activity regularization
activations_batch_averaged <-
tf$sqrt(tf$reduce_mean(tf$square(x), axis = 0L))
loss <- tf$reduce_sum(tf$multiply(reg_weights, activations_batch_averaged))
loss
}
# loss has 3 components: NLL, KL, and FNN
# otherwise, this is just normal TF2-style training
train_step_vae <- function(batch) {
with (tf$GradientTape(persistent = TRUE) %as% tape, {
code <- encoder(batch[[1]])
z <- reparameterize(code)
prediction <- decoder(z)
l_mse <- mse_loss(batch[[2]], prediction)
# see loss_false_nn in 2 previous posts
l_fnn <- loss_false_nn(code)
# KL divergence to a standard normal
l_kl <- -0.5 * k_mean(1 - k_square(z))
# overall loss is a weighted sum of all 3 components
loss <- l_mse + fnn_weight * l_fnn + kl_weight * l_kl
})
encoder_gradients <-
tape$gradient(loss, encoder$trainable_variables)
decoder_gradients <-
tape$gradient(loss, decoder$trainable_variables)
optimizer$apply_gradients(purrr::transpose(list(
encoder_gradients, encoder$trainable_variables
)))
optimizer$apply_gradients(purrr::transpose(list(
decoder_gradients, decoder$trainable_variables
)))
train_loss(loss)
train_mse(l_mse)
train_fnn(l_fnn)
train_kl(l_kl)
}
# wrap it all in autograph
training_loop_vae <- tf_function(autograph(function(ds_train) {
for (batch in ds_train) {
train_step_vae(batch)
#str(batch)
}
tf$print("Loss: ", train_loss$result())
tf$print("MSE: ", train_mse$result())
tf$print("FNN loss: ", train_fnn$result())
tf$print("KL loss: ", train_kl$result())
train_loss$reset_states()
train_mse$reset_states()
train_fnn$reset_states()
train_kl$reset_states()
}))
n_latent <- 10L
n_features <- 1
encoder <- vae_encoder_model(n_timesteps,
n_features,
n_latent)
decoder <- vae_decoder_model(n_timesteps,
n_features,
n_latent)
mse_loss <-
tf$keras$losses$MeanSquaredError(reduction = tf$keras$losses$Reduction$SUM)
train_loss <- tf$keras$metrics$Mean(name = 'train_loss')
train_fnn <- tf$keras$metrics$Mean(name = 'train_fnn')
train_mse <- tf$keras$metrics$Mean(name = 'train_mse')
train_kl <- tf$keras$metrics$Mean(name = 'train_kl')
fnn_multiplier <- 1 # default value used in nearly all cases (see text)
fnn_weight <- fnn_multiplier * nrow(x_train)/batch_size
kl_weight <- 1
optimizer <- optimizer_adam(learning_rate = 1e-3)
n_timesteps <- 120
batch_size <- 32
gen_timesteps <- function(x, n_timesteps) {
do.call(rbind,
purrr::map(seq_along(x),
function(i) {
start <- i
end <- i + n_timesteps - 1
out <- x[start:end]
out
})
) %>%
na.omit()
}
train <- gen_timesteps(roessler[1:(n/2)], 2 * n_timesteps)
test <- gen_timesteps(roessler[(n/2):n], 2 * n_timesteps)
dim(train) <- c(dim(train), 1)
dim(test) <- c(dim(test), 1)
x_train <- train[ , 1:n_timesteps, , drop = FALSE]
y_train <- train[ , (n_timesteps + 1):(2*n_timesteps), , drop = FALSE]
ds_train <- tensor_slices_dataset(list(x_train, y_train)) %>%
dataset_shuffle(nrow(x_train)) %>%
dataset_batch(batch_size)
x_test <- test[ , 1:n_timesteps, , drop = FALSE]
y_test <- test[ , (n_timesteps + 1):(2*n_timesteps), , drop = FALSE]
ds_test <- tensor_slices_dataset(list(x_test, y_test)) %>%
dataset_batch(nrow(x_test))
for (epoch in 1:100) {
cat("Epoch: ", epoch, " -----------\n")
training_loop_vae(ds_train)
test_batch <- as_iterator(ds_test) %>% iter_next()
encoded <- encoder(test_batch[[1]][1:1000])
test_var <- tf$math$reduce_variance(encoded, axis = 0L)
print(test_var %>% as.numeric() %>% round(5))
}
fails with the following message:
Error in py_call_impl(callable, call_args$unnamed, call_args$named) :
KeyError: 'The optimizer cannot recognize variable conv1d_transpose_24/kernel:0. This usually means you are trying to call the optimizer to update different parts of the model separately. Please call `optimizer.build(variables)` with the full list of trainable variables before the training loop or use legacy optimizer `tf.keras.optimizers.legacy.Adam.'
Run `reticulate::py_last_error()` for details.
Thanks.
Hello, I am testing code in https://blogs.rstudio.com/tensorflow/posts/2018-06-25-sunspots-lstm/
I am getting an error in this line:
cols <- map(coln, quo(sym(.)))
the error is:
Error in is_symbol(x): object '.' not found
Traceback:
1. map(coln, quo(sym(.)))
2. .f(.x[[i]], ...)
3. eval_tidy(~sym(.))
4. sym(.)
5. is_symbol(x)
I think is tidy related but I am not sure how I could fix it.
Could you give me an help?
Thanks.
Cheers
I was unable to load the example dataset in Tutorial: Text Classification example: . I'm using the keras library from rstudio:
imdb <- dataset_imdb(num_words = 10000)
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb.npz
17473536/17464789 [==============================] - 1s 0us/step
Error in py_call_impl(callable, dots$args, dots$keywords) :
ValueError: Object arrays cannot be loaded when allow_pickle=False
Thanks...Keith Erskine
ps: My environment is rstudio.cloud
> Sys.getenv()
_ /usr/bin/env
CLICOLOR_FORCE 1
DISPLAY :0
EDITOR vi
GIT_ASKPASS rpostback-askpass
HOME /home/rstudio-user
LANG C.UTF-8
LD_LIBRARY_PATH /opt/R/3.5.2/lib/R/lib::/lib:/usr/local/lib:/usr/lib/x86_64-linux-gnu:/usr/lib/jvm/java-8-openjdk-amd64/jre/lib/amd64/server
LD_PRELOAD /lib/x86_64-linux-gnu/libSegFault.so
LN_S ln -s
LOGNAME rstudio-user
MAIL /var/mail/rstudio-user
MAKE make
PAGER /usr/bin/pager
PATH /cloud/project/r-tensorflow/bin:/cloud/project/r-tensorflow/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/opt/R/3.5.2/lib/R/bin
PWD /
R_BROWSER xdg-open
R_BZIPCMD /bin/bzip2
R_DOC_DIR /opt/R/3.5.2/lib/R/doc
R_ENVIRON /etc/R/Renviron.site
R_GZIPCMD /bin/gzip
R_HOME /opt/R/3.5.2/lib/R
R_INCLUDE_DIR /opt/R/3.5.2/lib/R/include
R_LIBS_SITE
R_LIBS_USER ~/R/x86_64-pc-linux-gnu-library/3.5
R_PAPERSIZE letter
R_PDFVIEWER /usr/bin/xdg-open
R_PLATFORM x86_64-pc-linux-gnu
R_PRINTCMD /usr/bin/lpr
R_PROFILE /etc/R/Rprofile.site
R_RD4PDF times,inconsolata,hyper
R_SESSION_INITIALIZED PID=1683:NAME="reticulate"
R_SESSION_TMPDIR /tmp/Rtmp8rfxNG
R_SHARE_DIR /opt/R/3.5.2/lib/R/share
R_SYSTEM_ABI linux,gcc,gxx,gfortran,?
R_TEXI2DVICMD /usr/bin/texi2dvi
R_UNZIPCMD /usr/bin/unzip
R_ZIPCMD /usr/bin/zip
RETICULATE_REQUIRED_MODULE tensorflow
RMARKDOWN_MATHJAX_PATH /usr/lib/rstudio-server/resources/mathjax-26
RS_RPOSTBACK_PATH /usr/lib/rstudio-server/bin/rpostback
RSESSION_DIAGNOSTICS_ENABLED 1
RSESSION_DIAGNOSTICS_FILE /tmp/rsession-diagnostics-rstudio-user.log
RSESSION_PROFILE_OPTIONS
RSTUDIO 1
RSTUDIO_CONSOLE_COLOR 256
RSTUDIO_CONSOLE_WIDTH 126
RSTUDIO_DISABLE_PROJECT_SHARING 1
RSTUDIO_DISABLE_SECURE_DOWNLOAD_WARNING
1
RSTUDIO_HTTP_REFERER https://keith-erskine.rstudio.cloud/dcad4180e3f94b24b7ed14dce1a0667e/
RSTUDIO_PANDOC /usr/lib/rstudio-server/bin/pandoc
RSTUDIO_R_MODULE
RSTUDIO_R_PRELAUNCH_SCRIPT
RSTUDIO_R_VERSION_LABEL
RSTUDIO_SESSION_STREAM rstudio-user-d
RSTUDIO_USER_IDENTITY rstudio-user
RSTUDIO_VERSION 1.2.1206-2
RSTUDIO_WINUTILS bin/winutils
SED /bin/sed
SEGFAULT_OUTPUT_NAME /dev/fd/63
SEGFAULT_SIGNALS abrt segv
SHLVL 2
SSH_ASKPASS rpostback-askpass
TAR /bin/tar
TERM xterm-256color
USER rstudio-user
VIRTUAL_ENV /cloud/project/r-tensorflow
WORKON_HOME /cloud/project
Hi all,
I would like to puplish this article on R blog TensorFlow,
the link of my Github:
https://dsaada.github.io/CreditRiskTensorFlow/
Thanks,
[email protected]
David
Cc @skeydan
The master
branch of this repository will soon be renamed to main
, as part of a coordinated change across several GitHub organizations (including, but not limited to: tidyverse, r-lib, tidymodels, and sol-eng). We anticipate this will happen by the end of September 2021.
That will be preceded by a release of the usethis package, which will gain some functionality around detecting and adapting to a renamed default branch. There will also be a blog post at the time of this master
--> main
change.
The purpose of this issue is to:
message id: entire_lizard
On line 194, the link to "Time Series Analysis Example" should point to
https://tidymodels.github.io/rsample/articles/Applications/Time_Series.html
instead of
https://topepo.github.io/rsample/articles/Applications/Time_Series.html
as it is currently broken and I think this is new link. Thanks for this EXCELLENT post and keep up the good good good good work. 💯
In your article
Chollet & Allaire (2017, Dec. 20). TensorFlow for R: Time Series Forecasting with Recurrent Neural Networks. Retrieved from https://blogs.rstudio.com/tensorflow/posts/2017-12-20-time-series-forecasting-with-recurrent-neural-networks/
the code of the generator function appears to be untested! The generator function works on features in columns and samples in rows, but the data shown at the top and as explained/treated in the text is the other way around, samples in col and features in rows. The generator tries to access rows as samples, since there are only 16 Features, it immediately runs out of boundary
RGDS
hk
We should try to disable code that actually runs r code, see for instance:
Otherwise, when building the blog automatically, this might break or change inadvertently.
I would be pleased to propose my post entitled "Auto-Keras: An R easily accessible deep learning library" to be published in the TensorFlow for R Blog.
I have carefully followed the blog contribution guide detailed in https://blogs.rstudio.com/tensorflow/contributing.html .
You can find my proposed post in the URL https://github.com/jcrodriguez1989/tf_blog_autokeras .
I am at your entire disposal and open to any suggestions that will undoubtedly lead to improvements in my article.
Regards,
Juan Cruz Rodriguez
Hi @skeydan -
If/when the AI Blog moves to posit.co, please update the Google Analytics tracking code. Details on how to do that are in Confluence.
Thank you!
Sarah
I have 2 data files and i need to reconcile them using the RStudio. I cannot do this in Excel since the data length is too much. My current VB tool is manual and too much time consuming. So need to create a more quick tool. Can you please advise what is the best method and tool to be used? I am also open to use Python.
Data1 | ||||
---|---|---|---|---|
C1 | C2 | C3 | C4 | Total |
A | B | C | D | 5 |
A | B | C | D | 2 |
A | B | C | D1 | 1 |
A | B1 | C | D2 | 0 |
A | B2 | C | D | 0 |
A | B | C | D | 3 |
Data2 | |||||
---|---|---|---|---|---|
C5 | C6 | C7 | C7.1 | C8 | Total |
A | B | C | Cq | D | 5 |
A | B | C | Cq | D | 2 |
A | B | C | Cq | D1 | 1 |
A | B1 | C | Cq | D2 | 0 |
A | B2 | C | Cq | D | 0 |
A | B | C | Cq | D | 3 |
The Netlify step appears to run on every change, regardless of branch, resulting in no-op deployments in Netlify.
Should it have if: github.ref == 'refs/heads/main'
as a guard clause like the GitHub deploy?
https://github.com/rstudio/ai-blog/blob/main/.github/workflows/main.yml#L93
Hello, I am trying to get code from https://blogs.rstudio.com/tensorflow/posts/2018-07-30-attention-layer/ and trying to reproduce this example.
The following is my code:
reticulate::use_condaenv("tf-gpu", required = TRUE)
library(keras)
use_implementation("tensorflow")
library(tensorflow)
tfe_enable_eager_execution()
library(tfdatasets)
library(purrr)
library(stringr)
library(reshape2)
library(viridis)
library(ggplot2)
library(tibble)
filepath <- file.path("data", "nld.txt")
lines <- readLines(filepath, n = 10000)
sentences <- str_split(lines, "\t")
str(sentences)
space_before_punct <- function(sentence) {
str_replace_all(sentence, "([?.!])", " \\1")
}
replace_special_chars <- function(sentence) {
str_replace_all(sentence, "[^a-zA-Z?.!,¿]+", " ")
}
add_tokens <- function(sentence) {
paste0("<start> ", sentence, " <stop>")
}
add_tokens <- Vectorize(add_tokens, USE.NAMES = FALSE)
preprocess_sentence <- compose(add_tokens,
str_squish,
replace_special_chars,
space_before_punct)
word_pairs <- map(sentences, preprocess_sentence)
create_index <- function(sentences) {
unique_words <- sentences %>% unlist() %>% paste(collapse = " ") %>%
str_split(pattern = " ") %>% .[[1]] %>% unique() %>% sort()
index <- data.frame(
word = unique_words,
index = 1:length(unique_words),
stringsAsFactors = FALSE
) %>%
add_row(word = "<pad>",
index = 0,
.before = 1)
index
}
word2index <- function(word, index_df) {
index_df[index_df$word == word, "index"]
}
index2word <- function(index, index_df) {
index_df[index_df$index == index, "word"]
}
src_index <- create_index(map(word_pairs, ~ .[[1]]))
target_index <- create_index(map(word_pairs, ~ .[[2]]))
sentence2digits <- function(sentence, index_df) {
map((sentence %>% str_split(pattern = " "))[[1]], function(word)
word2index(word, index_df))
}
sentlist2diglist <- function(sentence_list, index_df) {
map(sentence_list, function(sentence)
sentence2digits(sentence, index_df))
}
src_diglist <- sentlist2diglist(map(word_pairs, ~ .[[1]]), src_index)
src_maxlen <- map(src_diglist, length) %>% unlist() %>% max()
src_matrix <- pad_sequences(src_diglist, maxlen = src_maxlen, padding = "post")
target_diglist <- sentlist2diglist(map(word_pairs, ~ .[[2]]), target_index)
target_maxlen <- map(target_diglist, length) %>% unlist() %>% max()
target_matrix <- pad_sequences(target_diglist, maxlen = target_maxlen, padding = "post")
train_indices <-
sample(nrow(src_matrix), size = nrow(src_matrix) * 0.8)
validation_indices <- setdiff(1:nrow(src_matrix), train_indices)
x_train <- src_matrix[train_indices, ]
y_train <- target_matrix[train_indices, ]
str(x_train)
str(y_train)
x_valid <- src_matrix[validation_indices, ]
y_valid <- target_matrix[validation_indices, ]
str(x_valid)
str(y_valid)
buffer_size <- nrow(x_train)
# just for convenience, so we may get a glimpse at translation
# performance during training
train_sentences <- sentences[train_indices]
validation_sentences <- sentences[validation_indices]
validation_sample <- sample(validation_sentences, 5)
str(train_sentences)
batch_size <- 32
embedding_dim <- 64
gru_units <- 256
src_vocab_size <- nrow(src_index)
target_vocab_size <- nrow(target_index)
train_dataset <-
tensor_slices_dataset(keras_array(list(x_train, y_train))) %>%
dataset_shuffle(buffer_size = buffer_size) %>%
dataset_batch(batch_size, drop_remainder = TRUE)
str(train_dataset)
validation_dataset <-
tensor_slices_dataset(keras_array(list(x_valid, y_valid))) %>%
dataset_shuffle(buffer_size = buffer_size) %>%
dataset_batch(batch_size, drop_remainder = TRUE)
str(validation_dataset)
attention_encoder <-
function(gru_units,
embedding_dim,
src_vocab_size,
name = NULL) {
keras_model_custom(name = name, function(self) {
self$embedding <-
layer_embedding(
input_dim = src_vocab_size,
output_dim = embedding_dim
)
self$gru <-
layer_gru(
units = gru_units,
return_sequences = TRUE,
return_state = TRUE
)
function(inputs, mask = NULL) {
x <- inputs[[1]]
hidden <- inputs[[2]]
x <- self$embedding(x)
c(output, state) %<-% self$gru(x, initial_state = hidden)
list(output, state)
}
})
}
attention_decoder <-
function(object,
gru_units,
embedding_dim,
target_vocab_size,
name = NULL) {
keras_model_custom(name = name, function(self) {
self$gru <-
layer_gru(
units = gru_units,
return_sequences = TRUE,
return_state = TRUE
)
self$embedding <-
layer_embedding(input_dim = target_vocab_size,
output_dim = embedding_dim)
gru_units <- gru_units
self$fc <- layer_dense(units = target_vocab_size)
self$W1 <- layer_dense(units = gru_units)
self$W2 <- layer_dense(units = gru_units)
self$V <- layer_dense(units = 1L)
function(inputs, mask = NULL) {
x <- inputs[[1]]
hidden <- inputs[[2]]
encoder_output <- inputs[[3]]
hidden_with_time_axis <- k_expand_dims(hidden, 2)
score <- self$V(k_tanh(self$W1(encoder_output) +
self$W2(hidden_with_time_axis)))
attention_weights <- k_softmax(score, axis = 2)
context_vector <- attention_weights * encoder_output
context_vector <- k_sum(context_vector, axis = 2)
x <- self$embedding(x)
x <- k_concatenate(list(k_expand_dims(context_vector, 2), x), axis = 3)
c(output, state) %<-% self$gru(x)
output <- k_reshape(output, c(-1, gru_units))
x <- self$fc(output)
list(x, state, attention_weights)
}
})
}
encoder <- attention_encoder(
gru_units = gru_units,
embedding_dim = embedding_dim,
src_vocab_size = src_vocab_size
)
decoder <- attention_decoder(
gru_units = gru_units,
embedding_dim = embedding_dim,
target_vocab_size = target_vocab_size
)
optimizer <- tf$compat$v1$train$AdamOptimizer()
cx_loss <- function(y_true, y_pred) {
mask <- ifelse(y_true == 0L, 0, 1)
loss <-
tf$nn$sparse_softmax_cross_entropy_with_logits(labels = y_true,
logits = y_pred) * mask
tf$reduce_mean(loss)
}
n_epochs <- 50
encoder_init_hidden <- k_zeros(c(batch_size, gru_units))
for (epoch in seq_len(n_epochs)) {
total_loss <- 0
iteration <- 0
iter <- make_iterator_one_shot(train_dataset)
until_out_of_range({
batch <- iterator_get_next(iter)
loss <- 0
x <- batch[[1]]
y <- batch[[2]]
iteration <- iteration + 1
with(tf$GradientTape() %as% tape, {
c(enc_output, enc_hidden) %<-% encoder(list(x, encoder_init_hidden))
dec_hidden <- enc_hidden
dec_input <-
k_expand_dims(rep(list(
word2index("<start>", target_index)
), batch_size))
for (t in seq_len(target_maxlen - 1)) {
c(preds, dec_hidden, weights) %<-%
decoder(list(dec_input, dec_hidden, enc_output))
loss <- loss + cx_loss(y[, t], preds)
dec_input <- k_expand_dims(y[, t])
}
})
total_loss <-
total_loss + loss / k_cast_to_floatx(dim(y)[2])
paste0(
"Batch loss (epoch/batch): ",
epoch,
"/",
iter,
": ",
(loss / k_cast_to_floatx(dim(y)[2])) %>%
as.double() %>% round(4),
"\n"
)
variables <- c(encoder$variables, decoder$variables)
gradients <- tape$gradient(loss, variables)
optimizer$apply_gradients(
purrr::transpose(list(gradients, variables)),
global_step = tf$train$get_or_create_global_step()
)
})
paste0(
"Total loss (epoch): ",
epoch,
": ",
(total_loss / k_cast_to_floatx(buffer_size)) %>%
as.double() %>% round(4),
"\n"
)
}
this code fails with the following error:
2020-02-12 12:48:30.175011: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_100.dll
Error: NotFoundError: Could not find valid device for node.
Node:{{node SparseSoftmaxCrossEntropyWithLogits}}
All kernels registered for op SparseSoftmaxCrossEntropyWithLogits :
device='CPU'; T in [DT_FLOAT]; Tlabels in [DT_INT32]
device='CPU'; T in [DT_FLOAT]; Tlabels in [DT_INT64]
device='CPU'; T in [DT_DOUBLE]; Tlabels in [DT_INT32]
device='CPU'; T in [DT_DOUBLE]; Tlabels in [DT_INT64]
device='CPU'; T in [DT_HALF]; Tlabels in [DT_INT32]
device='CPU'; T in [DT_HALF]; Tlabels in [DT_INT64]
device='GPU'; T in [DT_FLOAT]; Tlabels in [DT_INT32]
device='GPU'; T in [DT_FLOAT]; Tlabels in [DT_INT64]
device='GPU'; T in [DT_HALF]; Tlabels in [DT_INT32]
device='GPU'; T in [DT_HALF]; Tlabels in [DT_INT64]
[Op:SparseSoftmaxCrossEntropyWithLogits]
It is not clear to me what is the reason for this failure.
Do you get the same result?
Hi, I have just done writing a post for tensorflow blog (Loading Bert model from R).
Could you review it, please?
https://github.com/henry090/BERT-from-R
c(indices, target, segments) %<-% list(list(),list(),list())
Error in c(indices, target, segments) %<-% list(list(), list(), list()) :
could not find function "%<-%"
Can you help me, please?
Hi, Sigrid.
There are a lot of issues regarding the installation/configuration of TensorFlow by readers on the blog.
Would it be better if we have included reproducible examples at the end of the posts?
I can help to configure python+TF for recent posts on Google Colab in that case.
What do you think?
Hi, I really enjoy reading your blog 👍. As an author of {styler}, my eye is very trained on code styling and I noticed a few deviations from the tidyverse style guide in the last few blog entries I read, so I wondered if it's worth adding a section on coding style to the CONTRIBUTING.md (if you want to adhere to the tidyverse style). {styler} can style rmd via the addin or the CLI API, {knitr} can pretty-print the output (without source modification) when `tidy = 'styler' as described in the R markdown cookbook).
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.