Hi, im new to this forum. Currently preparing to enlist in the deep learning for coders, part 1.
Besides learning Python and having some practice with this, should one know how to use the libraries NumPy, Sci-Kit and MatPlotLib before venturing into the course???
Also would the course Computational Linear Algebra for Coders that are linked to be nice to have done beforehand, or should this wait until later??
Hello,
I am very glad to be part of the community. I would like to get more information about the VGG and SSD input size. Where could I find these informations?
At my work, we have products and when something does not quite fit into one of our categories, it goes into an others category. If I wanted to write a classifier for our images, would I have to figure out what categories are actually in the others category and then after predicting, if an image classifies as one of my made up categories, map it there? Or is it better to rely on a low prediction score against all the existing categories and that that as the āothersā classification?
Also, is there a limit on the number of categories one can try to classify?
iāve just started , but have not had to use any other libs. i also have gotten by with my existing knowledge of linear algebra, and iām not an expert.
I think the easiest way is to use additional unrelated categories and treat them as āothersā.
Actually, you can use existing imagenet 1000 categories as āothersā (you will need to download part of imagenet dataset to balance training)
There is a small probability that an āunknownā image will fall into your category instead of one in a thousand.
For more, you can read about Unsupervised learning
Basically, for each image you take activation valuesā from previous layers. You plot them (using t-SNE or similar) and calculate the distance between. Pictures which activation values are furthest away from the others probably do not belong to any category
AWS CloudFormation template used by the course isnāt respecting usage limits, after following instructions exactly on the course page and getting my limit increase. My p2.xlarge limit for region US-West-2 is set to 1, yet when I use the course CloudFormation template for US-West-2, I get the following error:
The following resource(s) failed to create: [FastaiNotebookInstance]. . Rollback requested by user. Physical ID: arn:aws:cloudformation:us-west-2:721532252063:stack/FastaiSageMakerStack/c7d4f520-75ba-11e9-a7fe-065f9ec2d268 Client Request Token: Console-CreateStack-395357cd-0396-4f3a-b165-86cb1532f0bc
13:08:19 UTC-0700 CREATE_FAILED AWS::SageMaker::NotebookInstance
FastaiNotebookInstance The account-level service limit āml.p2.xlarge for notebook instance usageā is 0 Instances, with current utilization of 0 Instances and a request delta of 1 Instances. Please contact AWS support to request an increase for this limit. (Service: AmazonSageMaker; Status Code: 400; Error Code: ResourceLimitExceeded; Request ID: d70dad85-9add-44f6-8d75-d9aac548f731)
Usage limits for the region in which the CloudFormation stack is being created:
Why does AWS claim I have a limit of 0 when it is clearly 1 in my EC2 instance limits for the region the CloudFormation stack is supposedly getting created in?
Sorry if this is the wrong location but I couldnāt post anywhere in the forums.
I am a bit confused between the 3 proposed courses on the Fast.ai website.
What is Introduction to Machine Learning for Coders! and how does it differ from Practical Deep Learning for Coders ?
Thanks a buch!
Machine Learning is about how to make computers work without explicit instructions
It includes techniques to handle data and build models. So you have xboost, random forest, ploting etc. The course is described here https://www.fast.ai/2018/09/26/ml-launch/
Deep Learning is a branch of ML where you focus particularly on neural networks. It is a really wide part of ML and is in the phase of intensive development now. Hence a separate course.
I think it doesnāt make much sense that one has to interact before asking a question, as most users only end up coming here once they have a question in mind. If it is due to preventing spam, a captcha would work well.
Speaking of questions, does anyone know why the image cleaner widget is not working in colab? It totally breaks my environment and I get disconnected, and then I need to restart everything.