I just treated others to an amazing post that I really enjoyed 
I think the difference is because of Jeremy’s philosophy of sharing the best with everyone, which inspires all of us 
1 Like
@helena I just wanted to say thank you again This had a big impact on how I write. I never knew you could not say everything. And that people would fill in the gaps and that it would make for better communication.
In fact, I tried to apply some of the points in my best story thus far. Small steps and a long way to go but making progress makes me happy
I signed up Very nice for practicing.
1 Like
Aaaand the promo coupon works
1 Like
Thread not updated in 19 days! Oh man 
Putting this here as I am not sure what other place would be good for it:
The biggest discovery of rewatching lec #1 and #2 - you can use cycle length to control overfitting. How cool is that!
3 Likes
This thread has wonderful advice at the very beginning but somehow is lost with other threads…
I think the thread died as I stopped sharing updates I am not sure really how interesting it is to people what I am up to these days. My strategy vs what I assumed was the correct approach at the beginning of part 1 v2 changed a lot.
I think the pre-edit version of this post sums up quite nicely how I view things now. Even the post edit version would do and has the nice benefit of being brief
Basically, with the absolutely great resources that we have access to thanks to fastai, all that remains is to practice. I absolutely suck at doing practical things, but through what I have written I am able to nudge myself very slowly towards being a better practitioner.
I think if we add to this continuing to ask ourselves the question that @mandroid6 asks in his post - how do I improve? how do I change my ways to become more efficient in learning and doing things, then we are on the right track.
I would also like to add that I am discovering that the ability to do things doesn’t lie in knowing more math, watching more non fastai lectures, or whatever else might have seemed like a no brainer initially but proved to be a dead end. Our capability to practice more lies in accumulating little tidbits that improve our efficiency (learn debugging, etc), learning simple things about the tools we use (ctags, plotting with matplotlib), being able to focus better on what is important and producing our own util files for experimenting with a particular problem (a method for drawing a bounding box, a method for returning parameter count in a model, etc).
I can only hope that this is the right approach but for now things move still quite slowly in relation to how much time I invest in learning, at least to my judgement. But maybe that is how it is meant to be.
So I ended up giving yet another update and rejuvenating this nearly dead thread thanks to you @ecdrid, after your post gave it a breath of fresh air
3 Likes
Me, on lesson 4 right now, searching forum, if someone tried doing NLP on Polish lang. Look at search results. O! Radek probably did something.
“Things that I will specifically NOT do though each is very tempting”
Great thing btw, list of things not to do. I’m going to incorporate something like this also.
I think the pre-edit version of this post sums up quite nicely how I view things now. Even the post edit version would do and has the nice benefit of being brief
Could you re-post this post in here? Limited access thing.
Hey sayko
If you go back to the post and click the little pencil symbol in the upper right hand corner it will show you the earlier version of the post.
I think it can be all summed up in a single phrase: write and read more code Quite amazing how versatile this approach is and applicable across a wide range of domains and technologies!
I have not done anything with Polish and NLP - have been considering to do so though. I don’t recall which sites those were (they offer rating) but one could scrape them and build a dataset for sentiment analysis, among other things. This could make for a fun project. Polish is quite unique in many ways (so many verb and noun forms!) so this could be quite an interesting and creative challenge. I am for now sticking to vision as it seems to be slightly more up my alley, at least for the time being. Maybe DeVISE will be something I spend more time on once I get to that lecture
Ah sorry only now read this part, somehow my brain skipped the last sentence and jumped into thinking you didn’t know how the pencil thingy works. Sorry!
Here it is. Read at your own risk BTW I would write all this stuff but I don’t mean to be giving advice to people in any greater way than I ma giving advice to myself. I often feel like the biggest newb around and the posts actually speak to my own deficiencies…
Anyhow, you asked for it yourself so don’t blame me!
I think if you are asking yourself this question, “how can I go about things in a better way, more efficiently”, like you are, you are already on the right track
Being here on the fastai forums you also seem to have found the best place I know off on the Internet for learning machine learning, deep learning and how to learn better in general and be a better programmer. It might seem that I am saying these things lightly but that is not the case - I do see it exactly this way.
There is one component though that you mention, that noone else can give you but you yourself. Nothing that I mention above will matter if you do not
This is something I need to focus on more myself. It is not a side dish that you eat once you know the theory and neither are those the veggies that you eat sporadically cause you know they are good for you. This is the main course. The secret ingredient. The breakfast, lunch and dinner of the folks who do awesome things (not speaking of myself here since I don’t have any claims to glory, but just listen to how Jeremy describes his own experience and what he mentions when speaking to the root of other people’s success).
So practice is what I will focus on more and more each day and at this point I don’t think that anything else is even remotely as important as practice.
I wrote this post some time ago on doing machine learning efficiently and same practices apply to practicing efficiently. There is also the whole set of ideas on how to be an efficient coder - long sessions of uninterrupted time that you devote to working on a problem (you do not need to put 80 hours into work a week, but even if you put only 10 hrs into smth, you will be exponentially better off if you do so without interruptions and in longer blocks of time - this is how the framework I use for work, Ruby on Rails, was born, out of 10 hrs a week and this seems to be the key to programmer efficiency, I am quite convinced 10 hrs of uninterrupted time > 40 hrs with interruptions), knowing your tools well, etc.
I bet you would enjoyed Cal Newport “Deep Work” book.
I’m also newly found believer of un-interupted blocks of time, I’m doing fast.ai course before work (waking up earlier) - it’s so quiet in the morning .
I struggle more with focus (so many interesting thing to explore!), but I think list of DONT’s will be a great help.
4 Likes
Added to my wishlist!