Comments (5)
Thank you very much for reporting this issue. Could you please point us to the code you are running or provide a reproducible test case?
from tensorflow_macos.
Hi, sure. The code is attached.
There's a lot of code related to pre-processing my specific data in the AutoClean class. Otherwise though, the main elements are in
- build_autoencoder()
- train()
- CustomCallback (callback related operations when epochs end)
Also, the setup of tensorFlow_macos is in the top part of the file, most of which I have been commenting in/out depending if I'm running on master tensorFlow, or tensorflow_macos.
Let me know if I can help any further.
from tensorflow_macos.
Hi @anna-tikhonova ,
I have modified my sample - this doesn't do much except demonstrate the problem. Hopefully it will be easier for you to test.
import sys
sys.path.append('../../')
import numpy as np
import warnings
from astropy.io.fits.verify import VerifyWarning
warnings.simplefilter('ignore', category=VerifyWarning)
import logging
import tensorflow as tf
from keras.optimizers import Adadelta
import tensorflow.keras.backend as K
from keras.layers import LeakyReLU, Input
from keras.models import Model
from keras.layers.convolutional import Conv2D, UpSampling2D, MaxPooling2D
# setup tensorflow_macos
# Import mlcompute module to use the optional set_mlc_device API for device selection with ML Compute.
EXECUTION_MODE = 'cpu'
from tensorflow.python.compiler.mlcompute import mlcompute
# Select CPU device.
mlcompute.set_mlc_device(device_name=EXECUTION_MODE) # Available options are 'cpu', 'gpu', and ‘any'.
# Turn off eager execution for GPU mode
if EXECUTION_MODE == 'gpu':
tf.config.run_functions_eagerly(False)
logging.basicConfig(filename='output.log', level=logging.DEBUG)
def recall_m(y_true, y_pred):
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))
recall = true_positives / (possible_positives + K.epsilon())
return recall
def precision_m(y_true, y_pred):
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))
precision = true_positives / (predicted_positives + K.epsilon())
return precision
def f1_m(y_true, y_pred):
precision = precision_m(y_true, y_pred)
recall = recall_m(y_true, y_pred)
return 2 * ((precision * recall) / (precision + recall + K.epsilon()))
class AutoClean:
def __init__(self, pretrained_model=None):
self.CNN_SIDE = 128
self.img_rows = self.CNN_SIDE
self.img_cols = self.CNN_SIDE
self.channels = 1
self.batch_size = 8
self.val_batch_size = 8
self.img_shape = (self.img_rows, self.img_cols, self.channels)
self.dataset_size = 5000
self.validation_size = 256
self.optimizer = Adadelta()
if pretrained_model:
self.aen = pretrained_model
else:
self.aen = self.build_autoencoder()
# self.aen = to_multi_gpu(self.aen, n_gpus=N_GPUS)
self.aen.compile(optimizer=self.optimizer, loss='binary_crossentropy', metrics=['accuracy',
f1_m,
precision_m,
recall_m])
def load_image_batch(self, n_samples):
"""
Gets a random batch of full and mask images from a directory with .npy files
:param n_samples:
:return:
"""
while True:
X_train_image = np.ones((n_samples,self.CNN_SIDE, self.CNN_SIDE,1))
X_train_mask = np.zeros((n_samples,self.CNN_SIDE, self.CNN_SIDE,1))
yield X_train_image, X_train_mask
def load_validation_batch(self, n_samples):
"""
Gets a the full validation batch of image and masks from a directory with .npy files
:param n_samples:
:return:
"""
while True:
X_vld_image = np.ones((n_samples,self.CNN_SIDE, self.CNN_SIDE,1))
X_vld_mask = np.ones((n_samples,self.CNN_SIDE, self.CNN_SIDE,1))
yield X_vld_image, X_vld_mask
def build_autoencoder(self):
depth = 128
input_img = Input(shape=self.img_shape)
# ENCODER
x = Conv2D(int(depth), kernel_size=3, strides=1, padding='same')(input_img)
x = LeakyReLU()(x)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(int(depth / 2), kernel_size=3, strides=1, padding='same')(x)
x = LeakyReLU()(x)
# x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(int(depth / 4), kernel_size=3, strides=1, padding='same')(x)
x = LeakyReLU()(x)
# x = MaxPooling2D((2, 2), padding='same')(x)
# x = Conv2D(int(depth/4), kernel_size=3, strides=1, padding='same')(x)
# x = LeakyReLU()(x)
# # x = MaxPooling2D((2, 2), padding='same')(x)
# # DECODER
# x = Conv2D(int(depth/4), kernel_size=3, strides=1, padding='same')(x) # (encoded)
# x = LeakyReLU()(x)
# # x = UpSampling2D((2, 2))(x)
x = Conv2D(int(depth / 4), kernel_size=3, strides=1, padding='same')(x)
x = LeakyReLU()(x)
# x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(int(depth / 2), kernel_size=3, strides=1, padding='same')(x)
x = LeakyReLU()(x)
# x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(int(depth), kernel_size=3, strides=1, padding='same')(x)
x = UpSampling2D(size=(2, 2))(x)
x = LeakyReLU()(x)
decoded = Conv2D(1, kernel_size=3, strides=1, activation='sigmoid', padding='same')(x)
autoencoder = Model(input_img, decoded)
# autoencoder = CustomModel(input_img, decoded)
autoencoder.summary()
return autoencoder
def train(self, n_epochs=100, batch_size=8):
self.batch_size = batch_size
self.val_batch_size = batch_size
self.aen.fit(
self.load_image_batch(n_samples=self.batch_size),
epochs=n_epochs,
steps_per_epoch=int(self.dataset_size / self.batch_size),
validation_data=self.load_validation_batch(n_samples=self.val_batch_size),
validation_steps=int(self.validation_size / self.val_batch_size)
)
if __name__ == '__main__':
aen = AutoClean(pretrained_model=None)
aen.train(n_epochs=4, batch_size=20)
Here is the CPU history showing the problem:
On standard TensorFlow all cores are in use, and training time is doubled accordingly. Hope this helps :)
from tensorflow_macos.
I have the same issue, even on much simpler case.
Hardware : MacBook Air M1 - 8GB / 512 GB
Very simple convnet on MNIST.
import numpy as np
import time
import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras.utils import to_categorical
#set cpu or gpu
from tensorflow.python.compiler.mlcompute import mlcompute
mlcompute.set_mlc_device(device_name='cpu')
mnist = tf.keras.datasets.mnist
(train_images,train_labels),(test_images,test_labels) = mnist.load_data()
train_images=train_images.reshape((60000,28,28,1))
train_images=train_images.astype('float32')/255
test_images=test_images.reshape((10000,28,28,1))
test_images=test_images.astype('float32')/255
train_labels = to_categorical(train_labels)
test_labels = to_categorical(test_labels)
model = tf.keras.models.Sequential()
model.add(layers.Conv2D(32,(3,3),activation = 'relu',input_shape=(28,28,1)))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(64,(3,3),activation = 'relu'))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(64,(3,3),activation = 'relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64,activation='relu'))
model.add(layers.Dense(10,activation='softmax'))
model.compile(optimizer = 'rmsprop',
loss = 'categorical_crossentropy',
metrics = ['accuracy'])
print("Start Learning with tensorflow.keras")
start = time.time()
history = model.fit(train_images,train_labels,epochs=5,batch_size=128)
print("Ran in {} seconds".format(time.time() - start))
test_loss, test_acc = model.evaluate(test_images,test_labels)
print('test_acc:',test_acc)
I get underutilization of the CPU (see the right side).
This is the same for MLP and LSTM models.
from tensorflow_macos.
Could be that in the case of M1, tensor flow is using only the 4 high-performance cores (and not the 4 efficiency cores) with tensorflowe, which would be Cores 4/5/6/8?, whilst Cores 1/2/3/7 are left out - but I can't be sure of this, so perhaps wait for a reply from somebody else.
In my case, I'm on an Intel Mac, and the code I had made full use of all cores on standard tensor flow, but only 50% of cores on this macOS fork.
from tensorflow_macos.
Related Issues (20)
- Using `model.set_weights()` yields incorrect behavior when MLC is enabled HOT 2
- Kernel is dying when I run the tensorflow.keras.models.Sequential.Fit() on Apple M1 HOT 2
- Apple M1 TensorFlow HOT 6
- Can't use SpectralNormalization wrapper layer.
- When can get new version basic on tf release version HOT 1
- Initialization issues and model.predict doesn't work HOT 6
- Error while running keras.sequential HOT 2
- Tensorflow with gpu HOT 2
- issues while doing model.fit HOT 1
- Apple M1: process finished with exit code 132 (interrupted by signal 4: sigill) HOT 8
- Transformer hugginface BERT model not working HOT 3
- installer script fail
- ImportError: dlopen( ... _pywrap_tensorflow_internal.so: mach-o, but wrong architecture HOT 2
- Request for comment: re: future of TF development on macOS. HOT 9
- `model.fit` makes the kernel crash when passing a `class_weight`
- op type not registered `NormalizeUTF8` initializing BERT HOT 1
- TransposeMLCBytes nullptr error during training process HOT 1
- TensorFlow Addons
- Error retrieving python version, or python executable /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 not version 3.8. Please specify a Python 3.8 executable with the --python option.
- Check failed: outputs_[index].tensor == nullptr
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from tensorflow_macos.