luisguiserrano / manning Goto Github PK

hi Luis, one stupid question, which python version are these exercises written? I keep having issues with py3.6.13. Many Thanks
Guan

import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'

above lines are needed to run on Macbook for "A graphical example", and possibly need to install xgboost as well...

Hi

I've found some typos in your book:

On Pg 50, you have writren:

Case 3: If the price of the house is lower than the price the model predicted, and the number
of rooms is positive:
• Add 1 cent to the price per room
• Subtract 1 cent to the base price.

However, in case 3, the number of rooms is negative.

For context, take a look at this image https://imgur.com/a/NT5szwl

An alternative would be to write the following code
plt.scatter(features,labels) plt.show()

In the notebook, you have:

def perceptron_trick(weights, bias, features, label, learning_rate = 0.01):
    pred = prediction(weights, bias, features)
    for i in range(len(weights)):
        weights[i] += (label-pred)*features[i]*learning_rate
        bias += (label-pred)*learning_rate
    return weights, bias

but in the book, the bias is not done for each iteration of the loop:

def perceptron_trick(weights, bias, features, label, learning_rate = 0.01):
    pred = prediction(weights, bias, features)
    for i in range(len(weights)):
        weights[i] += (label-pred)*features[i]*learning_rate
    bias += (label-pred)*learning_rate
    return weights, bias

Unfortunately, this leads to different numbers than those that are published in the book

The python files depend on turicreate , but this is not available natively on Windows.
For example running : Naive Bayes.ipynb gives you ModuleNotFoundError: No module named 'turicreate' but when you try to install via conda or pip -it fails ,as no support natively exists for windows 10 .

Hi @luisguiserrano,

I recently started reading GML, which is excellent, and wanted to run some of the examples however I am having issues getting them running on a new Macbook Pro with an M1 chip. I am using pyenv, poetry, and a terminal with Rosetta to try and install TuriCreate but am having no luck. Please let me know what other info you need to help with this issue.

Simple_trick is not offered as an option in the first linear regression function. In the second linear regression function, it is called, but the arguments is incorrect (no learning rate).

In the second definition of linear_regressin of this block, it reads

def linear_regression(features, labels, learning_rate=0.01, epochs = 1000):
    price_per_room = random.random()
    base_price = random.random()
    errors = []
    for i in range(epochs):
        predictions = features[0]*price_per_room+base_price
        errors.append(rmse(labels, predictions))
        i = random.randint(0, len(features)-1)

I suppose the predictions should be calculated on all features instead of just the fixed first feature.

        predictions = features[0]*price_per_room+base_price

This makes predictions a scalar instead of a vector. Then, the following rmse makes no sense.

Should it be like this?

        predictions = features*price_per_room+base_price

Was going through the code in the Ensemble chapter.

It appears sklearn externals six is going to be deprecated and eventually removed. Recommend installing "six" : https://pypi.org/project/six/ and the code will change from:
from sklearn.externals.six import StringIO
to:
from six import StringIO