Besides for a DL 中文 vocab, I just realized there is an interesting task is hidden here:
As fastai has not yet been extremely popular in China (but I believe it will be in near future) now, I don’t think people have the chance to come up any translation for fastai specific terms yet. For example, for dataset we have 数据集 as nearly common term for anyone to know, I don’t know how the Chinese academics would name DataBunch, or maybe only you and the fastai people have the right to determine its Chinese counterpart, or maybe no Chinese counterpart is necessary for DataBunch (but for crappify, I am sure we need a translation such as 垃圾化). What do you think?
Therefore, would you be interested to guide the translation style for fastai specific terms?
Well, your Chinese is beating me already, it takes me quite a while to come up this second one 残次化. “残”:(it feels like) missing, crippled, disabled, falling part; “次”:coming from the term “次品”, as product not meeting quality standard when inspected on manufacture line.
There could be more or better options but probably way beyond my vocab limit.
the two options above are fine with me, but I believe it’s the respected and beloved fastai creators’ pronunciation or announcement of its name really gives the life to it.
Do you think we should add a section in the wiki for fastai created terms? People can suggest options, and you make the final official call for settling the translation?
Previously, I tend to use the term model all the time referring everything from the architecture of a NN, to all the parameters as a whole undergoing training, to the finished model ready to predict.
After watch the videos a few times, there seems much better to differentiate model from learner in the following way:
model only refers to the neural network architecture, the framework, maybe including parameters before training
learner only refers to all parameters with certain architecture undergoing training to learn to do a task well
therefore, when we talk about overfitting and underfitting, we are referring to learner rather than model
Is this the preferred way of using learner and model for fastai world?
model contains the layers of the network and its weights/biases, and some status flags (e.g. trainable) - it’s the same in any DL framework.
Learner (and its subclasses) is a fastai “feature” that binds together data and model, optimizer, etc. and makes it easy to do almost everything a DL practitioner needs through a single interface.
So a good metaphor would be a bus - Learner is the driver console, model is perhaps the engine, and there are many other parts to the bus. And you Daniel are its driver that controls them all via that console that is Learner.
I’m glad I inspired you to finalise the last vocabulary. I also checked with BaiDu, the result for element-wise is 逐個, I find it is a bit inform. Yours is good.
Congratulations again for translating the v3 Part 1 with all the supplementary documents. You are a star.
The discussion with you are very helpful and encouraging, and I am glad you like 逐一,and I will update it soon.
Thanks again for your encouragement, but I am far from a star in our fastai community. I really love what fast.ai and this forum are building and sharing, and overwhelmed by the huge amount treasure generated here. I guess translation turns out to be the perfect way for me to calm down and take things in slowly and contribute back at the same time.
Also without the support and help from you and others on the forum, it will be much harder to keep going.