I created a custom layer in python so that I can feed the data directly.
but I noticed it runs extremely slow and the GPU usage is at most 1% ( the memory is allocated, i.e. I can see that when I run the script, it allocates 2100MB VRAM and terminating the training, frees around 1G.
I'm not sure if this is an expected behavior or I'm doing something wrong.
Here is the script I wrote (based on this former pr) :
import json
import caffe
import numpy as np
from random import shuffle
from PIL import Image
class MyDataLayer(caffe.Layer):
"""
This is a simple datalayer for training a network on CIFAR10.
"""
def setup(self, bottom, top):
self.top_names = ['data', 'label']
# === Read input parameters ===
params = eval(self.param_str)
# Check the paramameters for validity.
check_params(params)
# store input as class variables
self.batch_size = params['batch_size']
# Create a batch loader to load the images.
self.batch_loader = BatchLoader(params, None)
# === reshape tops ===
# since we use a fixed input image size, we can shape the data layer
# once. Else, we'd have to do it in the reshape call.
top[0].reshape(self.batch_size, 3, params['im_height'], params['im_width'])
# this is for our label, since we only have one label we set this to 1
top[1].reshape(self.batch_size, 1)
print_info("MyDataLayer", params)
def forward(self, bottom, top):
"""
Load data.
"""
for itt in range(self.batch_size):
# Use the batch loader to load the next image.
im, label = self.batch_loader.load_next_image()
# Add directly to the caffe data layer
top[0].data[itt, ...] = im
top[1].data[itt, ...] = label
def reshape(self, bottom, top):
"""
There is no need to reshape the data, since the input is of fixed size
(rows and columns)
"""
pass
def backward(self, top, propagate_down, bottom):
"""
These layers does not back propagate
"""
pass
class BatchLoader(object):
"""
This class abstracts away the loading of images.
Images can either be loaded singly, or in a batch. The latter is used for
the asyncronous data layer to preload batches while other processing is
performed.
labels:
the format is like :
png_data_batch_1/leptodactylus_pentadactylus_s_000004.png 6
png_data_batch_1/camion_s_000148.png 9
png_data_batch_1/tipper_truck_s_001250.png 9
"""
def __init__(self, params, result):
self.result = result
self.batch_size = params['batch_size']
self.image_root = params['image_root']
self.im_shape = [params['im_height'],params['im_width']]
# get list of images and their labels.
self.image_labels = params['label']
#getting the list of all image filenames along with their labels
self.imagelist = [line.rstrip('\n\r') for line in open(self.image_labels)]
self._cur = 0 # current image
# this class does some simple data-manipulations
self.transformer = SimpleTransformer()
print ("BatchLoader initialized with {} images".format(len(self.imagelist)))
def load_next_image(self):
"""
Load the next image in a batch.
"""
# Did we finish an epoch?
if self._cur == len(self.imagelist):
self._cur = 0
shuffle(self.imagelist)
# Load an image
image_and_label = self.imagelist[self._cur] # Get the image index
#read the image filename
image_file_name = image_and_label[0:-1]
#load the image
im = np.asarray(Image.open(self.image_root +'/'+image_file_name))
#im = scipy.misc.imresize(im, self.im_shape) # resize
# do a simple horizontal flip as data augmentation
flip = np.random.choice(2)*2-1
im = im[:, ::flip, :]
# Load and prepare ground truth
#read the label
label = image_and_label[-1]
#convert to onehot encoded vector
#fix: caffe automatically converts the label into one hot encoded vector. so we only need to simply use the decimal number (i.e. the plain label number)
#one_hot_label = np.eye(10)[label]
self._cur += 1
return self.transformer.preprocess(im), label
def check_params(params):
"""
A utility function to check the parameters for the data layers.
"""
required = ['batch_size', 'image_root', 'im_width', 'im_height', 'label']
for r in required:
assert r in params.keys(), 'Params must include {}'.format(r)
def print_info(name, params):
"""
Ouput some info regarding the class
"""
print ("{} initialized for split: {}, with bs: {}, im_shape: {}.".format(
name,
params['image_root'],
params['batch_size'],
params['im_height'],
params['im_width'],
params['label']))
class SimpleTransformer:
"""
SimpleTransformer is a simple class for preprocessing and deprocessing
images for caffe.
"""
def __init__(self, mean=[125.30, 123.05, 114.06]):
self.mean = np.array(mean, dtype=np.float32)
self.scale = 1.0
def set_mean(self, mean):
"""
Set the mean to subtract for centering the data.
"""
self.mean = mean
def set_scale(self, scale):
"""
Set the data scaling.
"""
self.scale = scale
def preprocess(self, im):
"""
preprocess() emulate the pre-processing occuring in the vgg16 caffe
prototxt.
"""
im = np.float32(im)
im = im[:, :, ::-1] # change to BGR
im -= self.mean
im *= self.scale
im = im.transpose((2, 0, 1))
return im
def deprocess(self, im):
"""
inverse of preprocess()
"""
im = im.transpose(1, 2, 0)
im /= self.scale
im += self.mean
im = im[:, :, ::-1] # change to RGB
return np.uint8(im)
And in my train_test.prototxt file I have :
name: "CIFAR10_SimpleTest_PythonLayer"
layer {
name: 'MyPythonLayer'
type: 'Python'
top: 'data'
top: 'label'
include {
phase: TRAIN
}
python_param {
#the python script filename
module: 'mypythonlayer'
#the class name
layer: 'MyDataLayer'
#needed parameters in json
param_str: '{"phase":"TRAIN", "batch_size":10, "im_height":32, "im_width":32, "image_root": "G:/Caffe/examples/cifar10/testbed/Train and Test using Pycaffe", "label": "G:/Caffe/examples/cifar10/testbed/Train and Test using Pycaffe/train_cifar10.txt"}'
}
}
layer {
name: 'MyPythonLayer'
type: 'Python'
top: 'data'
top: 'label'
include {
phase: TEST
}
python_param {
#the python script filename
module: 'mypythonlayer'
#the class name
layer: 'MyDataLayer'
#needed parameters in json
param_str: '{"phase":"TEST", "batch_size":10, "im_height":32, "im_width":32, "image_root": "G:/Caffe/examples/cifar10/testbed/Train and Test using Pycaffe", "label": "G:/Caffe/examples/cifar10/testbed/Train and Test using Pycaffe/test_cifar10.txt"}'
}
}
Whats wrong here?
Your data layer is not efficient enough and it takes most of the training time (you should try caffe time ... to get a more detailed profiling). At each forward pass you are waiting for the python layer to read batch_size images from disk one after the other. This can take forever.
You should consider using Multiprocessing to perform the reading at the background while the net is processing the previous batches: this should give you good CPU/GPU utilization.
See this example for multiprocessing python data layer.
Python layers are executed on CPU not the GPU so it's slow because things have to keep going between the CPU and GPU when training. That's also why you see low gpu usage because its waiting on the cpu to execute the python layer.
Related
I wrote a script using xgboost to predict soil class for a certain area using data from field and satellite images. The script as below:
`
rm(list=ls())
library(xgboost)
library(caret)
library(raster)
library(sp)
library(rgeos)
library(ggplot2)
setwd("G:/DATA")
data <- read.csv('96PointsClay02finalone.csv')
head(data)
summary(data)
dim(data)
ras <- stack("Allindices04TIFF.tif")
names(ras) <- c("b1", "b2", "b3", "b4", "b5", "b6", "b7", "b10", "b11","DEM",
"R1011", "SCI", "SAVI", "NDVI", "NDSI", "NDSandI", "MBSI",
"GSI", "GSAVI", "EVI", "DryBSI", "BIL", "BI","SRCI")
set.seed(27) # set seed for generating random data.
# createDataPartition() function from the caret package to split the original dataset into a training and testing set and split data into training (80%) and testing set (20%)
parts = createDataPartition(data$Clay, p = .8, list = F)
train = data[parts, ]
test = data[-parts, ]
#define predictor and response variables in training set
train_x = data.matrix(train[, -1])
train_y = train[,1]
#define predictor and response variables in testing set
test_x = data.matrix(test[, -1])
test_y = test[, 1]
#define final training and testing sets
xgb_train = xgb.DMatrix(data = train_x, label = train_y)
xgb_test = xgb.DMatrix(data = test_x, label = test_y)
#defining a watchlist
watchlist = list(train=xgb_train, test=xgb_test)
#fit XGBoost model and display training and testing data at each iteartion
model = xgb.train(data = xgb_train, max.depth = 3, watchlist=watchlist, nrounds = 100)
#define final model
model_xgboost = xgboost(data = xgb_train, max.depth = 3, nrounds = 86, verbose = 0)
summary(model_xgboost)
#use model to make predictions on test data
pred_y = predict(model_xgboost, xgb_test)
# performance metrics on the test data
mean((test_y - pred_y)^2) #mse - Mean Squared Error
caret::RMSE(test_y, pred_y) #rmse - Root Mean Squared Error
y_test_mean = mean(test_y)
rmseE<- function(error)
{
sqrt(mean(error^2))
}
y = test_y
yhat = pred_y
rmseresult=rmseE(y-yhat)
(r2 = R2(yhat , y, form = "traditional"))
cat('The R-square of the test data is ', round(r2,4), ' and the RMSE is ', round(rmseresult,4), '\n')
#use model to make predictions on satellite image
result <- predict(model_xgboost, ras[1:(nrow(ras)*ncol(ras))])
#create a result raster
res <- raster(ras)
#fill in results and add a "1" to them (to get back to initial class numbering! - see above "Prepare data" for more information)
res <- setValues(res,result+1)
#Save the output .tif file into saved directory
writeRaster(res, "xgbmodel_output", format = "GTiff", overwrite=T)
`
The script works well till it reachs
result <- predict(model_xgboost, ras[1:(nrow(ras)*ncol(ras))])
it takes some time then gives this error:
Error in predict.xgb.Booster(model_xgboost, ras[1:(nrow(ras) * ncol(ras))]) :
Feature names stored in `object` and `newdata` are different!
I realize that I am doing something wrong in that line. However, I do not know how to apply the xgboost model to a raster image that represents my study area.
It would be highly appreciated if someone give a hand, enlightened me, and helped me solve this problem....
My data as csv and raster image can be found here.
Finally, I got the reason for this error.
It was my mistake as the number of columns in the traning data was not the same as in the number of layers in the satellite image.
I would like to add information to my current dataset. At the moment, I have six-frame sequences in folders. The DataLoader reads all 6 and uses the first 3 for predicting the last 1/2/3 (depending on how many I tell him to). This is the function for the DataLoader.
class TrainFeeder(Dataset):
def init(self, data_set):
super(TrainFeeder, self).init()
self.input_data = data_set
#print(torch.cuda.current_device())
if torch.cuda.current_device() ==0:
print('There are total %d sequences in trainset' % len(self.input_data))
def getitem(self, index):
path = self.input_data[index]
imgs_path = sorted(glob.glob(path + '/*.png'))
imgs = []
for img_path in imgs_path:
img = cv2.imread(img_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img = cv2.resize(img, (256,448))
img = cv2.resize(img, (0, 0), fx=0.5, fy=0.5, interpolation=cv2.INTER_CUBIC) #has been 0.5 for official data, new is fx = 2.63 and fy = 2.84
img_tensor = ToTensor()(img).float()
imgs.append(img_tensor)
imgs = torch.stack(imgs, dim=0)
return imgs
def len(self):
return len(self.input_data)
Now I'd like to add one value to these images. It is a boolean, I have stored in a list in a .json in the same folder, like the six-frame-sequences. But I don't know how to add the values of the list in the .json to the tensor. Which dimension should I use? Will the system work at all, if I change the shape of the input?
The function getitem can return anything, so you can return a tuple instead of just images :
def __getitem__(self, index):
path = ...
# load your 6 images
imgs = torch.stack( ... )
# load your boolean metadata
metadata = load_json_data( ... )
# return them both
return (imgs, metadata)
You will need to make metadata a tensor before returning it, otherwise I expect that pytorch will complain about not being able to collate (i.e stack) them to make batches
"Will the system work" is a question only you can answer, since you did not provide the code of your ML model. I would bet on : "no but it won't require significant changes to work". Most likely you currently have a loop like
for imgs in dataloader:
# do some training
output = model(imgs)
...
And you will have to make it like
for imgs, metadata in dataloader:
# do some training
output = model(imgs)
...
I am trying to stack a few pre-trained models that I have through taking the last hidden layer of each model and then concatenating them together and then plugging them into a meta-learner model (e.g. XGBoost).
I am running into a big problem of having to process each image of my dataset multiple times since each base model requires a different processing method. This is causing my model to take a really long time to train and is infeasible. Is there any way to work past this?
For example:
model_1, processor_1 = pretrained_model(), pretrained_processor()
model_2, processor_2 = pretrained_model2(), pretrained_processor2()
for img in images:
input_1 = processor_1(img)
input_2 = processor_2(img)
out_1 = model_1(input_1)
out_2 = model_2(input_2)
torch.cat((out1,out2), dim=1) #concatenates hidden representations to feed into another model
Here'a recommendation if you want to process your images faster:
Note: I did not test this out
import torch
import torch.nn as nn
# Create a stack nn module
class StackedModel(nn.Module):
def __init__(self, model1, model2):
super(StackedModel, self).__init__()
self.model1 = model1
self.model2 = model2
def forward(self, imgs):
out_1 = model_1(input_1)
out_2 = model_2(input_2)
return torch.cat((out1, out2), dim=1)
# Init model
model = StackedModel(model1, model2)
# Try to stack and run in a larger batch assuming u have extra gpu space
stacked_preproc1 = []
stacked_preproc2 = []
max_batch_size = 16
total_output = []
for index, img in enumerate(images):
input_1 = processor_1(img)
input_2 = processor_2(img)
stacked_preproc1.append(input_1)
stakced_preproc2.appennd(input2)
if index % max_batch_size == 0:
stacked_preproc1 = torch.stack(stacked_preproc1)
stakced_preproc2 = torch.stack(stakced_preproc2)
else:
total_output.append(
model(stacked_preproc1, stacked_preproc2)
)
# Reset array
stacked_preproc1 = []
stakced_preproc2 = []
I'm using the coil-100 dataset which has images of 100 objects, 72 images per object taken from a fixed camera by turning the object 5 degrees per image. Following is the folder structure I'm using:
data/train/obj1/obj01_0.png, obj01_5.png ... obj01_355.png
.
.
data/train/obj85/obj85_0.png, obj85_5.png ... obj85_355.png
.
.
data/test/obj86/obj86_0.ong, obj86_5.png ... obj86_355.png
.
.
data/test/obj100/obj100_0.ong, obj100_5.png ... obj100_355.png
I have used the imageloader and dataloader classes. The train and test datasets loaded properly and I can print the class names.
train_path = 'data/train/'
test_path = 'data/test/'
data_transforms = {
transforms.Compose([
transforms.Resize(224, 224),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
train_data = torchvision.datasets.ImageFolder(
root=train_path,
transform= data_transforms
)
test_data = torchvision.datasets.ImageFolder(
root = test_path,
transform = data_transforms
)
train_loader = torch.utils.data.DataLoader(
train_data,
batch_size=None,
num_workers=1,
shuffle=False
)
test_loader = torch.utils.data.DataLoader(
test_data,
batch_size=None,
num_workers=1,
shuffle=False
)
print(len(train_data))
print(len(test_data))
classes = train_data.class_to_idx
print("detected classes: ", classes)
In my model I wish to pass every image through pretrained resnet and make a dataset from the output of resnet to feed into a biderectional LSTM.
For which I need to access the images by classname and index.
for ex. pre_resnet_train_data['obj01'][0] should be obj01_0.png and post_resnet_train_data['obj01'][0] should be the resnet output of obj01_0.png and so on.
I'm a beginner in Pytorch and for the past 2 days, I have read many tutorials and stackoverflow questions about creating a custom dataset class but couldn't figure out how to achieve what I want.
please help!
Assuming you only plan on running resent on the images once and save the output for later use, I suggest you write your own data set, derived from ImageFolder.
Save each resnet output at the same location as the image file with .pth extension.
class MyDataset(torchvision.datasets.ImageFolder):
def __init__(self, root, transform):
super(MyDataset, self).__init__(root, transform)
def __getitem__(self, index):
# override ImageFolder's method
"""
Args:
index (int): Index
Returns:
tuple: (sample, resnet, target) where target is class_index of the target class.
"""
path, target = self.samples[index]
sample = self.loader(path)
if self.transform is not None:
sample = self.transform(sample)
if self.target_transform is not None:
target = self.target_transform(target)
# this is where you load your resnet data
resnet_path = os.path.join(os.path.splitext(path)[0], '.pth') # replace image extension with .pth
resnet = torch.load(resnet_path) # load the stored features
return sample, resnet, target
I am getting a value error for parameters (not enough to unpack expected 2 got 1) I have a network I want to train:
def build(self):
numpy.random.seed(self.seed)
self.estimators.append(('standardize', StandardScaler))
self.estimators.append(('mlp', KerasClassifier(build_fn=self.build_fn, epochs=50, batch_size=5, verbose=0)))
self.pipeline = Pipeline(self.estimators)
Now if I want to fit the data to some values: say self.X, self.Y
self.model = self.pipeline.fit(self.X, self.Y, verbose=1)
I get
Traceback (most recent call last):
File "C:/Users/jaehan/PycharmProjects/cerebro/cerebro.py", line 257, in
<module>
model.run()
File "C:/Users/jaehan/PycharmProjects/cerebro/cerebro.py", line 138, in run
self.model = self.pipeline.fit(self.X, self.Y, verbose=1)
File "C:\Users\jaehan\AppData\Local\Continuum\anaconda3\envs\py36\lib\site-
packages\sklearn\pipeline.py", line 248, in fit
Xt, fit_params = self._fit(X, y, **fit_params)
File "C:\Users\jaehan\AppData\Local\Continuum\anaconda3\envs\py36\lib\site-
packages\sklearn\pipeline.py", line 197, in _fit
step, param = pname.split('__', 1)
ValueError: not enough values to unpack (expected 2, got 1)
Am I doing something wrong here? I was under the impression I could just run a fit and it would return a history object, which I could save and load at any time
I even tried...
self.pipeline.fit(self.X, self.Y)
Which throws...
AttributeError: 'numpy.ndarray' object has no attribute 'fit'
I have no idea what is going on here.
Full Code
class Cerebro:
def __init__(self):
self.model = None
self.build_fn = None
self.data = None
self.X = None
self.Y = None
#these three are for encoding string values to integer_encodings / one hot encodings
self.encoder = LabelEncoder()
self.encodings = {}
self.one_hot_encodings = {}
self.seed = numpy.random.seed(7) #this is to ensure we have reproducible results.
self.estimators = []
self.pipeline = None
self.kfold = StratifiedKFold(n_splits=10, shuffle=True, random_state=self.seed)
self.cross_validation_score = 0.0
def preprocess(self):
"""
This method will preprocess the dataset we want to train our network on.
Example:
import preproccessing
...
dataset, X, Y = preprocessing.main()
"""
self.data = pandas.read_csv('src_examples/hwtxn_final_for_influx.txt', sep='\t').values
self.X = numpy.delete(self.data, 13, axis=1)
self.Y = self.data[:, 13].astype(numpy.float16)
def build(self):
self.build_fn = self.base_model()
self.preprocess()
numpy.random.seed(self.seed)
self.estimators.append(('standardize', StandardScaler()))
self.estimators.append(('mlp', KerasClassifier(build_fn=self.build_fn, epochs=50, batch_size=5, verbose=0)))
self.pipeline = Pipeline(self.estimators)
def run(self):
"""This will actually take the pipeline (preprocessing standardization, model)
and fit it to our dataset (X, Y) (We don't need test/train since we are using stratified k fold cross val.)
Args:
None
Returns:
None
"""
# this is the 'model'
# self.pipeline
print(type(self.pipeline))
print(self.X.shape)
self.model = self.pipeline.fit(self.X, self.Y)
def load(self, fn):
"""This will load a saved model (history object)
Args:
fn (filename): represents saved model file
Returns:
model (pkl object): represents model
"""
return pickle.load(open(fn, 'rb'))
def save(self, fn):
"""This will save a model (history object)
Args:
fn (filename): represents a filename to save the model as
Returns:
None
"""
pickle.dump(self.model, open(fn, 'wb'))
def encode(self, vals, key):
""" This method will encode a list of values and take a key (representing column name, or index) to save
in the class object (self.encodings)
This will help us keep track of encodings we have for values we need to translate/decipher.
Args:
vals(np.array): array of values to encode
key(str): str representing the key used to encode this particular set of values
Returns:
transformed values (np.array) representing the encoded versions of values
"""
# int encoding for non int values
self.encodings[key] = self.encoder.fit_transform(vals)
return self.encoder.fit_transform(vals)
def decoder(self, vals, key):
"""This method will decode the integer_encodings for class variables. It will take vals which
represents a list of values to decode (i.e. [1,2,3] -- [apple, pear, orange])
It will also take a key (since every decoding has a corresponding encoding) to find which encoding
scheme to map to
Args:
vals(np.array) : array of values to decode
key(str) : string representing the key used for encoding the values (for decoding it)
Returns:
inverse transform of encoded values (np.array)
"""
# translate int encodings to original values (encoder._classes)
return self.encodings[key].inverse_transform(vals)
def cross_validate(self):
"""
This will perform a cross validation score using a stratified kfold method. (Think traditional Kfold but
with the values evenly distributed for each subsample)
Args:
None
Returns:
None
"""
self.cross_validation_score = cross_val_score(self.pipeline, self.X, self.Y, cv=self.kfold)
return self.cross_validation_score
#staticmethod
def base_model():
"""
This will return a base model for us to try. The good thing about this implementation is that
when we decide we want something more complex then all we have to do is define a class function and replace
the values in the build f(x)
Args:
None
Returns:
model (keras.models.Sequential): Keras based DNN Model
"""
# create model
model = Sequential()
model.add(Dense(60, input_dim=60, kernel_initializer='normal', activation='relu'))
model.add(Dense(1, kernel_initializer='normal', activation='sigmoid'))
# Compile model
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
return model
#staticmethod
def one_hot_encoder(int_encoding):
"""
This will take an integer encoding of string variables (traditional preprocessing step, will probably
move this to the preprocessing package.
Essential it returns a binary 'one hot' encoding of the values we wish to encode
Example
#Dataset Values
[apple, orange, pear]
#Integer Encoding
[1, 2, 3]
#One Hot Encoding
[[1, 0, 0]
[0, 1, 0]
[0, 0, 1]]
Args:
None
Returns:
Matrix (np.array): matrix representing one hot vectors for a class of values
"""
# we might not need this... so for now we will keep it static
return OneHotEncoder(sparse=False).fit_transform(int_encoding.reshape(len(int_encoding), 1))
if __name__ == '__main__':
# Step 1 is to initialize class (with seed == 7)
model = Cerebro()
model.build()
model.cross_validate()
print("Here are our estimators:\n {}".format(model.estimators))
print("Here is our pipeline:\n {}".format(model.pipeline))
model.run()
EDIT
The answer is that .fit() build_fn argument requires a function pointer and not the model itself.
IMHO I feel an error should be thrown for specifically that case.
This is due to the following line:
self.build_fn = self.base_model()
This should actually be:
self.build_fn = self.base_model
KerasClassifier requires a pointer to the function which creates the model, but by appending () at the end, you are assigning build_fn with the actual model, which is wrong.
Now in addition to above error, I would recommend checking the following lines in your code, which if not corrected will give error in future when you will use the code.
1) self.encodings[key] = self.encoder.fit_transform(vals)
Here you are assigning the transformed data to the encodings[key] not the model. So when you do this:-
self.encodings[key].inverse_transform(vals)
It makes no sense to call inverse_transform() on the transformed data.
inverse_transform() is a method of scikit-learn transformers. But self.encodings[key] will give out a ndarray, because you have saved the output array from fit_transform().
2) Something similar to 2 is also happening with one_hot_encoder()
The error "AttributeError: 'numpy.ndarray' object has no attribute 'fit'" seems related to 1 and 2.