Another treat! Early access to Intro To Machine Learning videos


(Eternal) #394

Hey Eric,

Will you be completing this?


(swetha Godi) #397

It’ll be really helpful if these parameters are explained in a notebook Or any link to the resources explaining the same should work. Since there are no pre-requisites mentioned, I just started the course with an open mind.


(Aditya) #398

Question on pdp-plots

How to interpret this pdp-plot (its from some other dataset)( a lot of -ve lines)

Dataset is basically on re-sale of house prices


(nok) #399

I have finished the first 7 lecture. Starting from Lecture 8, the course shift the focus to deep learning. Is there overlap with the deeplearining part1 v2?


(Jeremy Howard) #400

Excellent! :slight_smile: Note that you are looking at a preview release of the course so some minor bumps are to be expected. We’re all here to help as best we can though! BTW, there’s some excellent tips in #part1v2-beg showing how you can create a pull request (PR) to contribute fixes to the courses and libraries. If you’re interested in trying that out, we’d be delighted to take a look at a PR that adds this kind of info to the notebooks. Many students have found that creating PRs as you learn is a great way to test and solidify your own understanding, whilst helping out the next students that come along!


(Jeremy Demlow) #401
  1. I am not to sure about one so i will not try to answer that because i don’t want to send you in the wrong direction

  2. m.fit(df, y) * Essentially this is saying fit my data from the data.frame on my dependent variable
    df = dataframe
    y = the dependent variable that you are interested in

  3. This is giving you RMSE: (Root Mean Square Error (RMSE) is the standard deviation of the residuals (prediction errors). Residuals are a measure of how far from the regression line data points are; RMSE is a measure of how spread out these residuals are. In other words, it tells you how concentrated the data is around the line of best fit. Root mean square error is commonly used in climatology, forecasting, and regression analysis to verify experimental results.) here we are using a rdf so it is a little different, but that is how i think about it.
    Scoring is another way to understand the validity of your predictions
    ??I am not sure what hsattr is doing.

hope that helps and i look forward to seeing who can help even further as i still have questions myself


(Jeremy Howard) #402

X is the independent variables, y is the dependent. These are used nearly universally in ML so you’ll be seeing them a lot!

It’s not that important frankly; you’ll learn later in the course that OOB scores are only provided if you ask for them, so we can’t print them out if they’re not there. Googling for hasattr will give details on the function itself - it’s part of standard python.


(Jeremy Demlow) #403

Awesome and thank you for the response, I got a little confused because
"def fit(self, X, y, sample_weight=None):" i didn’t realize that was just the data.frame

I will be googling hasattr() and looking at the source code here shortly.

Jeremy,

I do have two questions for you if you don’t mind.

1.) i am currently taking the DL course that you have created and now that i found this I want to take the ML and DL course. Do you recommend taking them side by side or independently

2.) Also does the fast.ai library need to be run on an instance or can it be used similar to any other package? I would like to be able to use the fast.ai libraries functionality on class projects with out having to be on a server and when i try to import fast.ai i get the following: (I have no issues when i am using it on AWS)

ModuleNotFoundError Traceback (most recent call last)
in ()
1
2 # This file contains all the main external libs we’ll use
----> 3 from fastai.imports import *
4 from fastai.transforms import *
5 from fastai.conv_learner import *

ModuleNotFoundError: No module named ‘fastai’

I have googled this and everything on google is from the forums or related to your course or all mention be apart of an instance and if that is the case it’s okay I will just have to pay the money to run the instance. I am currently in transition to Paperspace as AWS bill was very expensive.

I am so glad that I ended up finding this course from the forums. I am attending LSU and going through their analytic program and I am using your DL course as an independent study, being able to learn practical and cutting edge techniques that can be used right away is priceless. With the added bonus of then going in and understanding how they work is a learning method i am loving.

Again thank you for all your help,

Jeremy


(Ahmad Arib) #404

Try to answer your question based on Jeremy answer that I read so many times, haha.

  1. He suggests taking it side by side, I start from DL, up until lesson 3, keep hear Jeremy suggest taking ML course, switch to ML video until lesson 10, now stuck in ML project academic paper using Random Forest. Will continue ML until finish then back to DL to finish I guess.

  2. what do you mean by “similar to any other package?” and “without having to be on a server”, is that mean you would like to run it on your local machine? PC or laptop?
    I run fastai di cluster server, similar to AWS and paperspace for DL related.
    While just using my laptop for ML related, you just need to install it like this page tell you https://github.com/fastai/fastai
    And I run fastai both in cluster and laptop using Linux Ubuntu.


(Jeremy Demlow) #406

ahmadarib,

I would like to be able to use the fastai library in the anaconda environment, but on my local machine not using a instance on paperspace.

I attempted to use the resource that you gave me, but to no avail. This could be because I wasn’t using the right command prompt. I have a windows 10 & 64 bit computer.

I had to use git bash to get the copy of the repo and used python install setup.py and I thought that did the trick inside the anaconda environment.

At the end of the day i can always use Paperspace to do this, but I’d prefer not to be charged to be able to experiment with ML course.


(Rishaan S Patel) #407

No module found usually refers to the package not being installed on your computer. When you write from fastai.imports import * you are basically telling the computer to go grab fastai library from your computer and load it up because you are about to use. But if the package isn’t installed, then you get that error. Python has multiple ways of installing packages like pip and anaconda, so you might want to look into those. Running pip install bcolz in your terminal should install the package.


(Ahmad Arib) #408

Ahh, my bad, I point you the github link that only could work out of the box if you’re using Linux machine.
And just like a week ago I installing fastai in my girlfriend laptop that also use windows 10 to learn ML videos and practicing it locally, and success, don’t give up yet, lot’s of people could do what you want.
I follow this link How to set up Windows 10 for fast.ai
Get stuck in first option, succeed in second option.
However, thread owner said it’s already obsolete, you need to check out this link also Howto: installation on Windows

Just follow the link above, you need to use Anaconda Prompt to execute everything.
And bcolz is one of them, please follow that 2 link first until sucess, using Anaconda Prompt.
And let us know if you still encounter problem like no module named bcolz, or etc.
Because when you already do right step by step, that error should not be appear at all.


(swetha Godi) #409

That’s an awesome idea :slight_smile: I shall try that out.
Thanks for sharing the information.


(swetha Godi) #410

Thanks for sharing :slight_smile:


(Jeremy Demlow) #411

Howto: installation on Windows “For Context Here Is the Link Used” Make Sure you’re running as an administrator

I made the mistake of not running as an admin the first through and first time I did this the conda env update did work and it didn’t get stuck, but as an admin it took a little longer so be prepared to wait a couple mins and don’t panic

Followed the rest of the sets and was able to get jupyter to run and the clone worked as you can see

The issue is when I run the from Fast.ai.imports isn’t there and bcolz issue anymore though :slight_smile:


(Aditya) #412

If I am not wrong then your current directory isn’t correct?

Try doing a cd into courses and then either ml or dl and then it should work?


(Jeremy Demlow) #413

I have been trying all types of combination of these using it as an admin and a non-admin to see if that was the case this is what i did and still got the same result with no module named ‘fastai’. Should my path have Anaconda3?

At this point i can get to the juypter notebook the library works, but the bash commands aren’t working.

I believe that this is because jupyternotebook in running power shell and this could be why this isn’t working so i may need to use powershell code in the terminal to get the bulldozer data. I will work on finding out the equivalents to BASH -> Powershell and see if this is the case. I would of course love to use the BASH commands if that is possible, but I don’t know how to go about doing that

AWESOME!! Got it working, but it’s very slow compared to paperspace this was very much worth doing and was successful


(Aditya) #414

Its Windows, so we have to use !dir {PATH}

In Linux ls works…


(Aditya) #415

(Sorry for spamming the forum)

Regarding the pdp plots,

How can we have -ve values for our prediction of the sale-prices even though RF’s hasn’t been trained on any?? (as far as i have read about RF’s they can only predict the values alike on what they have been trained)?

Thanks for replying…
@ramesh @jamesrequa @radek (sorry)(can’t remember more)


#416

I don’t think random forests can predict something they have not been trained on :wink: