This should be a simple task: Download a model saved in tensorflow_hub format, load using tensorflow_hub, and use..
This is the model I am trying to use (simCLR stored in Google Cloud): https://console.cloud.google.com/storage/browser/simclr-checkpoints/simclrv2/pretrained/r50_1x_sk0;tab=objects?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22))&prefix=&forceOnObjectsSortingFiltering=false
I downloaded the /hub folder as they say, using
gsutil -m cp -r \
"gs://simclr-checkpoints/simclrv2/pretrained/r50_1x_sk0/hub" \
.
The /hub folder contains the files:
/saved_model.pb
/tfhub_module.pb
/variables/variables.index
/variables/variables.data-00000-of-00001
So far so good.
Now in python3, tensorflow2, tensorflow_hub 0.12 I run the following code:
import numpy as np
import tensorflow as tf
import tensorflow_hub as hub
path_to_hub = '/home/my_name/my_path/simclr/hub'
# Attempt 1
m = tf.keras.models.Sequential([hub.KerasLayer(path_to_hub, input_shape=(224,224,3))])
# Attempt 2
m = tf.keras.models.Sequential(hub.KerasLayer(hubmod))
m.build(input_shape=[None,224,224,3])
# Attempt 3
m = hub.KerasLayer(hub.load(hubmod))
# Toy Data Test
X = np.random.random((1,244,244,3)).astype(np.float32)
y = m.predict(X)
None of these 3 options to load the hub model work, with the following errors:
Attempt 1 :
ValueError: Error when checking input: expected keras_layer_2_input to have shape (224, 224, 3) but got array with shape (244, 244, 3)
Attempt 2:
tensorflow.python.framework.errors_impl.UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node sequential_3/keras_layer_3/StatefulPartitionedCall/base_model/conv2d/Conv2D}}]] [Op:__inference_keras_scratch_graph_46402]
Function call stack:
keras_scratch_graph
Attempt 3:
ValueError: Expected a string, got <tensorflow.python.training.tracking.tracking.AutoTrackable object at 0x7fa71c7a2dd0>
These 3 attempts are all code taken from tensorflow_hub tutorials and are repeated in other answers in stackoverflow, but none works, and I don't know how to continue from those error messages.
Appreciate any help, thanks.
Update 1:
Same issues happen if I try with this ResNet50 hub/
https://storage.cloud.google.com/simclr-gcs/checkpoints/ResNet50_1x.zip
As #Frightera pointed out, there was an error with the input shapes. Also the error on "Attempt 2" was solved by allowing for memory growth on the selected GPU. "Attempt 3" still does not work, but at least there are two methods for loading and using a model saved in /hub format:
import numpy as np
import tensorflow as tf
import tensorflow_hub as hub
gpus = tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_visible_devices(gpus[0], 'GPU')
tf.config.experimental.set_memory_growth(gpus[0], True)
hubmod = 'https://tfhub.dev/google/imagenet/mobilenet_v2_035_96/feature_vector/5'
# Alternative 1 - Works!
m = tf.keras.models.Sequential([hub.KerasLayer(hubmod, input_shape=(96,96,3))])
print(m.summary())
# Alternative 2 - Works!
m = tf.keras.models.Sequential(hub.KerasLayer(hubmod))
m.build(input_shape=[None, 96,96,3])
print(m.summary())
# Alternative 3 - Doesnt work
#m = hub.KerasLayer(hub.load(hubmod))
#m.build(input_shape=[None, 96,96,3])
#print(m.summary())
# Test
X = np.random.random((1,96,96,3)).astype(np.float32)
y = m.predict(X)
print(y.shape)
Related
I am currently writing a script to augment a dataset for me using tf.keras (code given below). I'm pretty new to tf and data augmentation so I've been following a tutorial (https://blog.devgenius.io/data-augmentation-programming-e9a4703198be) pretty religiously. Despite this, I've been running into a lot of errors when I try to actually apply the ImageDataGenerator object to the image I'm loading. Specifically, I keep getting this error:
Exception has occurred: FileNotFoundError
[Errno 2] No such file or directory: '/home/kai/SURF22/yolov5/data/sc_google_aug/aug_0_3413.png'
File "/home/kai/SURF22/yolov5/data_augmentation", line 45, in <module>
for batch in idg.flow(aug_array,
It seems like tf can't find the image I want it to augment but I have no idea why because I load the image and input it as an array like the tutorial does. I tried inputting the absolute file path to the image instead one time but then I got a "string to float" error. Basically, I have no idea what is wrong and no one else seems to be getting this error when applying a for loop to .flow(). If anyone has advice on what could be going wrong I'd really appreciate it!
# images folder directory
folder_dir = "/home/kai/SURF22/yolov5/data/"
# initialize count
i = 0
for image in os.listdir(folder_dir + "prelim_data/sc_google_trans"):
# open the image
img = Image.open(folder_dir + "prelim_data/sc_google_trans/" + image)
# make copy of image to augment
# want to preserve original image
aug_img = img.copy()
# define an ImageDataGenerator object
idg = ImageDataGenerator(horizontal_flip=True,
vertical_flip=True,
rotation_range=360,
brightness_range=[0.2, 1.0],
shear_range=45)
# aug_img = load_img(folder_dir + "prelim_data/sc_google_trans/0.png")
# reshape image to a 4D array to be used with keras flow function
aug_array = img_to_array(aug_img)
aug_array = aug_array.reshape((1,) + aug_array.shape)
# augment image
for batch in idg.flow(aug_array,
batch_size=1,
save_to_dir='/home/kai/SURF22/yolov5/data/sc_google_aug',
save_prefix='aug',
save_format='png'):
i += 1
if i > 3:
break
The following code predicts the p of the binomial distribution by using beta as prior. Somehow, sometimes, I get meaningless results (acceptance rate = 0). When I write the same logic with pymc3, I have no issue.
I couldn't see what I am missing here.
import numpy as np
import tensorflow as tf
import tensorflow_probability as tfp
import edward2 as ed
from pymc3.stats import hpd
import numpy as np
import seaborn
import matplotlib.pyplot as plt
p_true = .15
N = [10, 100, 1000]
successN = np.random.binomial(p=p_true, n=N)
print(N)
print(successN)
def beta_binomial(N):
p = ed.Beta(
concentration1=tf.ones( len(N) ),
concentration0=tf.ones( len(N) ),
name='p'
)
return ed.Binomial(total_count=N, probs=p, name='obs')
log_joint = ed.make_log_joint_fn(beta_binomial)
def target_log_prob_fn(p):
return log_joint(N=N, p=p, obs=successN)
#kernel = tfp.mcmc.HamiltonianMonteCarlo(
# target_log_prob_fn=target_log_prob_fn,
# step_size=0.01,
# num_leapfrog_steps=5)
kernel = tfp.mcmc.NoUTurnSampler(
target_log_prob_fn=target_log_prob_fn,
step_size=.01
)
trace, kernel_results = tfp.mcmc.sample_chain(
num_results=1000,
kernel=kernel,
num_burnin_steps=500,
current_state=[
tf.random.uniform(( len(N) ,))
],
trace_fn=(lambda current_state, kernel_results: kernel_results),
return_final_kernel_results=False)
p, = trace
p = p.numpy()
print(p.shape)
print('acceptance rate ', np.mean(kernel_results.is_accepted))
def printSummary(name, v):
print(name, v.shape)
print(np.mean(v, axis=0))
print(hpd(v))
printSummary('p', p)
for data in p.T:
print(data.shape)
seaborn.distplot(data, kde=False)
plt.savefig('p.png')
Libraries:
pip install -U pip
pip install -e git+https://github.com/google/edward2.git#4a8ed9f5b1dac0190867c48e816168f9f28b5129#egg=edward2
pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-2.0.0-cp37-cp37m-manylinux2010_x86_64.whl#egg=tensorflow
pip install tensorflow-probability
Sometimes I see the following (when acceptance rate=0):
And, sometimes I see the following (when acceptance rate>.9):
When I get unstable results in Bayesian inference (I use mc-stan, but it's also using NUTS), it's usually because either the priors and likelihood are mis-specified, or the hyperparameters are not good for the problem.
That first graph shows that the sampler never moved away from the initial guess at the answers (hence the 0 acceptance rate). It also worries me that the green distribution seems to be right on 0. The beta(1,1) has positive probability at 0 but a p=0 might be an unstable solution here? (as in, the sampler may not be able to calculate the derivative at that point and returns a NaN, so doesn't know where to sample next?? Complete guess there).
Can you force the initial condition to be 0 and see if that always creates a failed sampling?
Other than that, I would try tweaking the hyperparameters, such as step size, number of iterations, etc...
Also, you may want to simplify the example by only using one N. Might help you diagnose. Good luck!
random.uniform's maxval default value is None. I changed it to 1, the result became stable.
random.uniform(( len(N) ,), minval=0, maxval=1)
I am trying to read text in a file Shakespear.txt line by line using
tensorflow TextLineDataset. Split the words in a line and write the words in another file txt.txt one word per line. Here is my code
import tensorflow as tf
tf.enable_eager_execution()
BATCH_SIZE=2
#from tensorflow.keras.model import Sequential
dataset_in_lines=tf.data.TextLineDataset("Shakespear.txt")
dataset=dataset_in_lines.map(lambda string: tf.string_split([string]).values)
with open("txt.txt","w") as f:
for k in dataset.take(2):
for x in k:
f.write("\n".join(x))
When i run it it gives the error: Cannot iterate over a scalar tensor
in the f.write line. Please help me figure out the issue
It will be helpful if you could share the shakespear.txt file, but based on your error, it seems liek it is receiving the tensor not the actual value.
So, you first need to get the value from tensor k, you can use k.numpy().
Replace for x in k: with for x in k.numpy():
Let us know if it works.
I found a better way, replace dataset=dataset_in_lines.map(lambda string:tf.string_split([string]).values) with tokenizer.tokenize. The following code achieves the objective(see https://www.tensorflow.org/tutorials/load_data/text for more details)
import tensorflow as tf
tf.enable_eager_execution()
import tensorflow_datasets as tfds
tokenizer = tfds.features.text.Tokenizer()
dataset_in_lines=tf.data.TextLineDataset("Shakespear.txt")
vocabulary_set = set()
for x in dataset_in_lines:
k=tokenizer.tokenize(x.numpy())
vocabulary_set.update(k)
with open("txt.txt","w") as f:
for x in vocabulary_set:
f.write(x+"\n")
I downloaded the fasttext.cc vectors of 1.5gb, I used example code spaCy examples vectors_fast_text. I executed the following command in the terminal:
python config/vectors_fast_text.py vectors_loc data/vectors/wiki.pt.vec
After a few minutes with the processor at 100%, I received the following text:
class colspan 0.32231358
What happens from here? How can I export these vectors elsewhere, such as for example with my AWS S3 training templates?
I modified the example script, to load the existing data of my language, read the file word2vec and at the end write all the content in a folder (this folder needs to exist).
Follow vectors_fast_text.py:
[LANGUAGE] = example: "pt"
[FILE_WORD2VEC] = "./data/word2vec.txt"
from __future__ import unicode_literals
import plac
import numpy
import spacy
from spacy.language import Language
#plac.annotations()
def main():
nlp = spacy.load('[LANGUAGE]')
with open("[FILE_WORD2VEC]", 'rb') as file_:
header = file_.readline()
nr_row, nr_dim = header.split()
nlp.vocab.reset_vectors(width=int(nr_dim))
count = 0
for line in file_:
count += 1
line = line.rstrip().decode('utf8')
pieces = line.rsplit(' ', int(nr_dim))
word = pieces[0]
print("{} - {}".format(count, word))
vector = numpy.asarray([float(v) for v in pieces[1:]], dtype='f')
nlp.vocab.set_vector(word, vector) # add the vectors to the vocab
nlp.to_disk("./models/new_nlp/")
if __name__ == '__main__':
plac.call(main)
Type in the terminal:
python vectors_fast_text.py
It will take about 10 minutes to finish, depending on the size of the word2vec file. In the script I made the print of the word, so that you can follow.
After that, you must type in the terminal:
python -m spacy package ./models/new_nlp/ ./my_models/
python setup.py sdist
And then you will have a "zip" file.
pip install /path/to/pt_example_model-1.0.0.tar.gz
A detailed tutorial can be found on the spaCy website:
https://spacy.io/usage/training
Trying debug statements in Python/tensorflow1.0 using jupyter , but does not get any output printed from tf.Print
Thought sess.run(during training in below code) should have evaluated db1 tensor and print output which did not happen
However db1.eval in evaluate phase , printing entire tensor X with out "message X:".
def combine_inputs(X):
db1=tf.Print(X,[X],message='X:')
return (tf.matmul(X, W) + b,db1)
<<training code>>
_,summary=sess.run([train_op,merged_summaries])
## merged_summaries tensor triggers combine_inputs function. There are
## other tensor functions/coding in between , not giving entire code to keep
## it simple; code works as expected except tf.Print
<<evaluate code>>
print(db1.eval())
Confused on following
a) Why tf.Print is not printing during sess.run during training?
b) Why explicit db1.eval is necessary , expected tf.Print to trigger with
sess.run. If eval is required , could copy tensor X in my code to db1
and evaluate it with out tf.Print. Correct?
Tried going through other questions (like below one). Suggested to implement memory_util or predefined function. As learner could not understand why tf.Print does not work in my scenario
If anyone encountered similar issues , please assist. Thanks!
Similar question in stackoverflow
According to the documentation, tf.Print prints to standard error (as of version 1.1), and it's not compatible with jupyter notebook. That's why you can't see any output.
Check here:
https://www.tensorflow.org/api_docs/python/tf/Print
You can check the terminal where you launched the jupyter notebook to see the message.
import tensorflow as tf
tf.InteractiveSession()
a = tf.constant(1)
b = tf.constant(2)
opt = a + b
opt = tf.Print(opt, [opt], message="1 + 2 = ")
opt.eval()
In the terminal, I can see:
2018-01-02 23:38:07.691808: I tensorflow/core/kernels/logging_ops.cc:79] 1 + 2 = [3]