I am using PyTorch lightning, so lightning control GPU/CPU assignments and in
return I get easy multi GPU support for training.
I would like to create an embedding that does not fit in the GPU memory.
fit_in_cpu = torch.nn.Embedding(too_big_for_GPU, embedding_dim)
Then when I select the subset for a batch, send it to the GPU
GPU_tensor = embedding(idx)
How do I do this in Pytorch Lightning?
Lightning will send anything that is registered as a model parameter to GPU, i.e: weights of layers (anything in torch.nn.*) and variables registered using torch.nn.parameter.Parameter.
However if you want to declare something in CPU and then on runtime move it to GPU you can go 2 ways:
Create the too_big_for_GPU inside the __init__ without registering it as a model parameter (using torch.zeros or torch.randn or any other init function). Then move it to the GPU on the forward pass
class MyModule(pl.LightningModule):
def __init__():
self.too_big_for_GPU = torch.zeros(4, 1000, 1000, 1000)
def forward(self, x):
# Move tensor to same GPU as x and operate with it
y = self.too_big_for_GPU.to(x.device) * x**2
return y
Create the too_big_for_GPU which will be created by default in CPU and then you would need to move it to GPU
class MyModule(pl.LightningModule):
def forward(self, x):
# Create the tensor on the fly and move it to x GPU
too_big_for_GPU = torch.zeros(4, 1000, 1000, 1000).to(x.device)
# Operate with it
y = too_big_for_GPU * x**2
return y
I saved train and validation set as tfrecord file. Inference gives input images and returns logits. loss and accuracy compute loss and accuracy as well. Using this code network is trained well(train set accuracy increases and loss decreases). But accuracy of validation set is almost fixed. By tensorboard, I found that computing accuracy of validation set create new graph that doesn't use the main graph's weights. How can I predict accuracy on validation set simultaneously?
def run_training():
train_images,train_labels = read_and_decode_tfrecord_train(train_data_path)
val_images,val_labels = read_and_decode_tfrecord_validation(validation_data_path)
train_images = tf.cast(train_images,tf.float32)/255.
val_images = tf.cast(train_images,tf.float32)/255.
batch_Xs,batch_Ys=tf.train.shuffle_batch([train_images,train_labels],batch_size=500,capacity=500,min_after_dequeue=100)
batch_xs,batch_ys=tf.train.shuffle_batch([val_images,val_labels],batch_size=500,capacity=500,min_after_dequeue=100)
logits=inference(batch_Xs,1)
total_loss = loss(logits,batch_Ys)
train_op = training(total_loss,learning_rate=LEARNING_RATE)
accuracy = evaluation(logits,batch_Ys)
val_logits=inference(batch_xs,1)
val_accuracy = evaluation(val_logits,batch_ys)
saver = tf.train.Saver(tf.all_variables(), max_to_keep=4,)
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
tf.train.start_queue_runners(sess=sess)
for i in range(NUM_ITER):
_,loss_value,acc=sess.run([train_op,total_loss,accuracy])
if i%10==0:
val_acc,testing_summary_accuracy=sess.run([val_accuracy,testing_summary])
print 'Iteration:',i, ' Loss:',loss_value,' Train Accuracy:',acc,' Validation Accuracy:',v
using psychopy ver 1.81.03 on a mac I want to draw a polygon (e.g. a triangle) on top of an image.
So far, my image stays always on top and thus hides the polygon, no matter the order I put them in. This also stays true if I have the polygon start a frame later than the image.
e.g. see inn the code below (created with the Builder before compiling) how both a blue square and a red triangle are supposed to start at frame 0, but when you run it the blue square always covers the red triangle!?
Is there a way to have the polygon on top? Do I somehow need to merge the image and polygon before drawing them?
Thank you so much for your help!!
Sebastian
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
This experiment was created using PsychoPy2 Experiment Builder (v1.81.03), Sun Jan 18 20:44:26 2015
If you publish work using this script please cite the relevant PsychoPy publications
Peirce, JW (2007) PsychoPy - Psychophysics software in Python. Journal of Neuroscience Methods, 162(1-2), 8-13.
Peirce, JW (2009) Generating stimuli for neuroscience using PsychoPy. Frontiers in Neuroinformatics, 2:10. doi: 10.3389/neuro.11.010.2008
"""
from __future__ import division # so that 1/3=0.333 instead of 1/3=0
from psychopy import visual, core, data, event, logging, sound, gui
from psychopy.constants import * # things like STARTED, FINISHED
import numpy as np # whole numpy lib is available, prepend 'np.'
from numpy import sin, cos, tan, log, log10, pi, average, sqrt, std, deg2rad, rad2deg, linspace, asarray
from numpy.random import random, randint, normal, shuffle
import os # handy system and path functions
# Ensure that relative paths start from the same directory as this script
_thisDir = os.path.dirname(os.path.abspath(__file__))
os.chdir(_thisDir)
# Store info about the experiment session
expName = u'test_triangle_over_square' # from the Builder filename that created this script
expInfo = {'participant':'', 'session':'001'}
dlg = gui.DlgFromDict(dictionary=expInfo, title=expName)
if dlg.OK == False: core.quit() # user pressed cancel
expInfo['date'] = data.getDateStr() # add a simple timestamp
expInfo['expName'] = expName
# Data file name stem = absolute path + name; later add .psyexp, .csv, .log, etc
filename = _thisDir + os.sep + 'data/%s_%s_%s' %(expInfo['participant'], expName, expInfo['date'])
# An ExperimentHandler isn't essential but helps with data saving
thisExp = data.ExperimentHandler(name=expName, version='',
extraInfo=expInfo, runtimeInfo=None,
originPath=None,
savePickle=True, saveWideText=True,
dataFileName=filename)
#save a log file for detail verbose info
logFile = logging.LogFile(filename+'.log', level=logging.EXP)
logging.console.setLevel(logging.WARNING) # this outputs to the screen, not a file
endExpNow = False # flag for 'escape' or other condition => quit the exp
# Start Code - component code to be run before the window creation
# Setup the Window
win = visual.Window(size=(1280, 800), fullscr=True, screen=0, allowGUI=False, allowStencil=False,
monitor='testMonitor', color=[0,0,0], colorSpace='rgb',
blendMode='avg', useFBO=True,
)
# store frame rate of monitor if we can measure it successfully
expInfo['frameRate']=win.getActualFrameRate()
if expInfo['frameRate']!=None:
frameDur = 1.0/round(expInfo['frameRate'])
else:
frameDur = 1.0/60.0 # couldn't get a reliable measure so guess
# Initialize components for Routine "trial"
trialClock = core.Clock()
ISI = core.StaticPeriod(win=win, screenHz=expInfo['frameRate'], name='ISI')
square = visual.ImageStim(win=win, name='square',units='pix',
image=None, mask=None,
ori=0, pos=[0, 0], size=[200, 200],
color=u'blue', colorSpace='rgb', opacity=1,
flipHoriz=False, flipVert=False,
texRes=128, interpolate=True, depth=-1.0)
polygon = visual.ShapeStim(win=win, name='polygon',units='pix',
vertices = [[-[200, 300][0]/2.0,-[200, 300][1]/2.0], [+[200, 300][0]/2.0,-[200, 300][1]/2.0], [0,[200, 300][1]/2.0]],
ori=0, pos=[0, 0],
lineWidth=1, lineColor=[1,1,1], lineColorSpace='rgb',
fillColor=u'red', fillColorSpace='rgb',
opacity=1,interpolate=True)
# Create some handy timers
globalClock = core.Clock() # to track the time since experiment started
routineTimer = core.CountdownTimer() # to track time remaining of each (non-slip) routine
#------Prepare to start Routine "trial"-------
t = 0
trialClock.reset() # clock
frameN = -1
# update component parameters for each repeat
# keep track of which components have finished
trialComponents = []
trialComponents.append(ISI)
trialComponents.append(square)
trialComponents.append(polygon)
for thisComponent in trialComponents:
if hasattr(thisComponent, 'status'):
thisComponent.status = NOT_STARTED
#-------Start Routine "trial"-------
continueRoutine = True
while continueRoutine:
# get current time
t = trialClock.getTime()
frameN = frameN + 1 # number of completed frames (so 0 is the first frame)
# update/draw components on each frame
# *square* updates
if frameN >= 0 and square.status == NOT_STARTED:
# keep track of start time/frame for later
square.tStart = t # underestimates by a little under one frame
square.frameNStart = frameN # exact frame index
square.setAutoDraw(True)
# *polygon* updates
if frameN >= 0 and polygon.status == NOT_STARTED:
# keep track of start time/frame for later
polygon.tStart = t # underestimates by a little under one frame
polygon.frameNStart = frameN # exact frame index
polygon.setAutoDraw(True)
# *ISI* period
if t >= 0.0 and ISI.status == NOT_STARTED:
# keep track of start time/frame for later
ISI.tStart = t # underestimates by a little under one frame
ISI.frameNStart = frameN # exact frame index
ISI.start(0.5)
elif ISI.status == STARTED: #one frame should pass before updating params and completing
ISI.complete() #finish the static period
# check if all components have finished
if not continueRoutine: # a component has requested a forced-end of Routine
routineTimer.reset() # if we abort early the non-slip timer needs reset
break
continueRoutine = False # will revert to True if at least one component still running
for thisComponent in trialComponents:
if hasattr(thisComponent, "status") and thisComponent.status != FINISHED:
continueRoutine = True
break # at least one component has not yet finished
# check for quit (the Esc key)
if endExpNow or event.getKeys(keyList=["escape"]):
core.quit()
# refresh the screen
if continueRoutine: # don't flip if this routine is over or we'll get a blank screen
win.flip()
else: # this Routine was not non-slip safe so reset non-slip timer
routineTimer.reset()
#-------Ending Routine "trial"-------
for thisComponent in trialComponents:
if hasattr(thisComponent, "setAutoDraw"):
thisComponent.setAutoDraw(False)
win.close()
core.quit()
As per Jonas' comment above, PsychoPy uses a layering system in which subsequent stimuli are drawn on top of previous stimuli (as in his code examples).
In the graphical Builder environment, drawing order is represented by the vertical order of stimulus components: stimuli at the top are drawn first, and ones lower down are progressively layered upon them.
You can change the order of stimulus components by right-clicking on them and selecting "Move up", "move down", etc as required.
Sebastian, has, however, identified a bug here, in that the intended drawing order is not honoured between ImageStim and ShapeStim components. As a work-around, you might be able to replace your ShapeStim with a bitmap representation, displayed using an ImageStim. Multiple ImageStims should draw correctly (as do multiple ShapeStims). To get it to draw correctly on top of another image, be sure to save it as a .png file, which supports transparency. That way, only the actual shape will be drawn on top, as its background pixels can be set to be transparent and will not mask the the underlying image.
For a long-term solution, I've added your issue as a bug report to the PsychoPy GitHub project here:
https://github.com/psychopy/psychopy/issues/795
It turned out to be a bug in the Polygon component in Builder.
This is fixed in the upcoming release (1.82.00). The changes needed to make the fix can be seen at
https://github.com/psychopy/psychopy/commit/af1af9a7a85cee9b4ec8ad5e2ff1f03140bd1a36
which you can add to your own installation if you like.
cheers,
Jon
I’m trying to animate a Mayavi pipeline volume:
src = mlab.pipeline.volume(mlab.pipeline.scalar_field(data),vmin=.1*np.max(data),vmax=.2*np.max(data))
that is combined in the pipeline by another dataset represented as a cut plane.
However, I can’t get the volume visualization to update - only the first frame shows up. The animation is stepping through the data correctly (I get different values of the np.max(data[t]) below) but nothing in the visualization changes.
My understanding is that mlab_source_set should re-render correctly, and there’s nothing on the web anywhere that describes this (as far as I can tell).
The animation looks like:
#mlab.show
#mlab.animate(delay=250,ui=True)
def anim(src,data,tax,fig):
"""Animate."""
t = 0
nt = len(tax)
while 1:
vmin = .1*np.max(data[t])
vmax = .2*np.max(data[t])
print 'animation t = ',tax[t],', max = ',np.max(data[t])
src.mlab_source.set(scalar = mlab.pipeline.scalar_field(data[t]), vmin=vmin,vmax=vmax)
t = mod(t+1,nt)
yield
Any thoughts?
So I installed python 2.7.8 and the latest pygame, but I can't seem to get image.load () to find my image file, despite several attempts to rename and recheck the spelling. The image and script that use it are in the same directory. Has anyone run into similar issues?
self.src_image = pygame.image.load
Is the line in question. image is used as a parameter that is filled later with a specific file name.
Here is some context:
import pygame, math, sys, os
from pygame.locals import *
screen = pygame.display.set_mode((1024, 786))
clock = pygame.time.Clock()
class WitchSprite(pygame.sprite.Sprite):
speed = 10
acceleration = .4
def __init__(self, image, position):
pygame.sprite.Sprite.__init__(self)
self.src_image = pygame.image.load(os.path.join(image))
self.postion = position
self.speed = self.direction = 0
self.k_left = self.k_right = self.k_down = self.k_up = 0
First, check that your image can even be loaded. Use pygame.image.get_extended() to check if it will load image formats. It should return True.
The pygame docs say that you should "use os.path.join() for compatibility." Try to load your image using os and see if that takes care of your problem.
ex: "asurf = pygame.image.load(os.path.join('data', 'bla.png'))"