EDIMAR FERNANDES_201933940004_RELATÓRIO - TOPICO-5_APRENDIZAGEM EM CONJUNTO.pdf

I think there's a confusing typo in "Exercise 4: Threshold calibration": The curve is assigned a label="ROC curve" but I think that should be label="Precision-recall curve".

dfs = pd.read_csv('data.csv')
dfs
dfs.set_value(0, 'a', 10)
dfs.to_csv('data.csv', index=False) # Don't export the row index`

if this is run you get following error:

OSError: File b'data.csv' does not exist

I cannot find the data.csv anywhere in the repo either.

Some questions and suggestions came to mind when I read about the gradient descent method:

• In section Gradient Descent, I find the formulation of the exponential decay of the learning rate a little bit odd. I would suggest expressing \eta_s in terms of \eta_0 instead.

• In section Stochastic Gradient Descent (SGD), I believe it would be better to start the index at i=0, otherwise it would make more sense to divide by n+1 when averaging the individual losses. Same goes for the other two sums in that part.

• Furthermore, the "incremental gradient" method does look a lot like the SAG method described here instead of the incremental aggregated gradient (IAG) method from this paper, which I found confusing. I also found the SAGA Algorithm. Maybe adding some of these references will be helpful to other students.

• Another suggestion would be to change "random i" to "if i = i_s" and adding "with i_s randomly chosen per iteration".

The names of neural networks slides contain |. This seems allowed on macOS, but it is an invalid character in Windows and Linux file names. It prevents Windows and Linux users cloning or updating this repository.

Pulling from upstreams...
remote: Counting objects: 177, done.
remote: Compressing objects: 100% (10/10), done.
remote: Total 177 (delta 4), reused 1 (delta 1), pack-reused 166
Receiving objects: 100% (177/177), 11.75 MiB | 582.00 KiB/s, done.
Resolving deltas: 100% (10/10), completed with 1 local object.
From https://github.com/joaquinvanschoren/ML-course
 * branch            master     -> FETCH_HEAD
   66e8297..3162f05  master     -> upstream/master
error: unable to create file N1 | Introduction.ipynb: Invalid argument
error: unable to create file N2 | Artificial Neuron.ipynb: Invalid argument
error: unable to create file N3 | Perceptron Classifier.ipynb: Invalid argument
error: unable to create file N4 | MLP.ipynb: Invalid argument
error: unable to create file N5 | MLP Classifier.ipynb: Invalid argument
error: unable to create file N6 | Optimization and Regularization.ipynb: Invalid argument
error: unable to create file N7 | Convolutional Networks.ipynb: Invalid argument
Updating 66e8297..3162f05
error: unable to create file N8 | Recurrent Networks.ipynb: Invalid argument

Some packages are cut off out of the PDF.

Example:

In section "Neural networks: evaluation and optimization", after "E.g. Gradient descent:", the formula is missing a partial derivative symbol and, more importantly, I guess there should be a minus in order to move towards the minimum.

Running this code the process gets interrupted.

import openml as oml
import numpy as np
import matplotlib.pyplot as plt
import sklearn

# Download Streetview data. Takes a while the first time.
SVHN = oml.datasets.get_dataset(41081)
X, y, cats, attrs = SVHN.get_data(dataset_format='array',
    target=SVHN.default_target_attribute)

Process finished with exit code 137 (interrupted by signal 9: SIGKILL)
I have 12gig of memory free, can't figure out why I am getting this interrupt.

Some LaTeX functions are not being rendered correctly on the website. This seems to primarily affect the formula for accuracy, precision, etc.

Hi, the solutions to lab 2a and 7a are not available yet.
They are not listed on the home page /intro.html.

Could you put these online?

In Tutorial 4 there is an error

print("SVM component: {}".format(pipe.named_steps['svm']f))

The f should not be there

I'm getting this error when I try to run the code which should load the devnagari dataset:

This error is pretty generic, so searching for for a fix is hard. Does anybody know the cause of this error?

• I could not immediately find the idea behind the Gini impurity index on the internet. The following derivation helped me understand the intuition a little bit better:
  
  The idea is that this captures "how often a randomly selected element is labeled incorrectly if the label is chosen randomly according to the actual distribution (in a leaf)".
• The definition of information gain, it is unclear to me what X_i is exactly. I would have expected Gain(X, i) and |X| in the denominator of the fraction. Would that make sense? Furthermore, am I correct that this l=1 to L sum loops over what some call the levels of this feature?

When I was installing openML on windows, I had several issues. As it turns out, one of the packages included, netifaces, requires Visual C++ Build Tools to be installed. It would be helpful for future students with this issue if you include this information somewhere. Even if it is a very common program to have installed, I personally had recently emptied my hard disk so I did not have it installed. pip does not clearly state this package is required.

In FDM_Challenge_Part3 it is confusing that the 'Gabor family' and 'Responses' both plot 10 * 10 = 100 images. This gives the impression that len(kernel) == 100. In reality len(kernel) == 225.

You're probably already aware, but in a lot of places in the html view of the pages there are dollar signs around the formulas.

For example:

It seems to happen when double dollar signs ($$) are used in the notebooks to typeset LaTeX over a full line. For some reason the html view then sometimes (not always) treats it as inline LaTeX instead and treats the formula as if it was only surrounded with single dollar signs. Since the result is correct in the notebooks and PDFs I'm not sure if there is an easy way to resolve this. Luckily it also doesn't affect readability too much so this isn't a very high priority issue.

Finally, the issue seems to happen in both Firefox and Chrome.

In the lab solutions, the preamble is still imported, which causes an error (and it's overkill).
Update the imports with only the required packages, as in the orginal lab notebooks.

I usually sync my fork when the solutions to the labs are published but the labs 3 do not seem to be released yet.

They are available via the website.