bloominstituteoftechnology / ds-unit-4-sprint-2-neural-networks Goto Github PK

From @KeriKalmbach:

We really need to re-engineer module 3's assignment. I know we work with arrays, so the feature names aren’t shared. But, it makes sense that they’ll need the feature names to make a meaningful interpretation of the data and to meet the data engineering assignment requirement.

• Replace the Module 3 Assignment Regression Data with NYC Rent or another dataset.

From @KeriKalmbach:

When either of the pages/notebooks (links below) are opened directly into google colab from github, the file name changes from the correct 4.2.1 to an incorrect 4.3.1
Are there two versions of the file? Are updates not getting loaded into colab?

If the file/notebook is downloaded from a zipped repo download and then uploaded, the correct number remains in the title (no bug).

https://github.com/LambdaSchool/DS-Unit-4-Sprint-2-Neural-Networks/blob/master/module1-Intro-to-Neural-Networks/LS_DS_421_Intro_to_NN_Lecture.ipynb
or
https://github.com/LambdaSchool/DS-Unit-4-Sprint-2-Neural-Networks/blob/master/module1-Intro-to-Neural-Networks/LS_DS_421_Intro_to_NN_Assignment.ipynb

In module2-backpropagation/LS_DS_422_Gradient_Descent_Backprop_Lecture.ipynb the sidmoid derivative function (sigmoidPrime) is calculated incorrectly.
It currently is:

def sigmoidPrime(self, s):
     return s * (1 - s) 

and it should be:

def sigmoidPrime(self, s):
     sx = sigmoid(s)
     return sx * (1 - sx)

This class function can be found under the header 'Update Weights Based on Gradient'

The following "sentence" is not grammatically correct and fixing it in various ways changes the meaning of the sentence in various ways. What should this sentence say?

"Make sure to have your final layer have as many nodes as the number of classes that you want to predict."

In the module one lecture notebook, the cell:

Types of Layers:

There are three main types of neuron layers in a typical NN topology:

is followed by a cell that describes the input/visible layer. There is no description of the other types.

LS_DS_433_Keras_Lecture.ipynb

pulling on colab or pulling on local or pulling on aws results in different resulting files...

"but let us simply the problem for now"

In the "Put it all together" section from the lecture for module 1, there's the following code;

# Weighted sum of inputs / weights
weighted_sum = np.dot(inputs, weights)
# Activate!
activated_output = sigmoid(weighted_sum)
# Cac error
error = correct_outputs - activated_output
adjustments = error * sigmoid_derivate(activated_output)

The adjustments should be adjustments = error * sigmoid_derivate(weighted_sum). Intuitively, this is because we want the derivative of our activation function at the points defined by the weighted sum, not at the points after activation. It is the activation function we're taking the derivative of, after all.

The NeuralNetwork class constructed throughout module 2 applies sigmoidPrime to the activated output.

self.z2_delta = self.z2_error * self.sigmoidPrime(self.activated_hidden)

should be

self.z2_delta = self.z2_error * self.sigmoidPrime(self.hidden_sum)

and

self.o_delta = self.o_error * self.sigmoidPrime(o)

should be

self.o_delta = self.o_error * self.sigmoidPrime(self.output_sum)

By sheer coincidence, this error cancels out with the error in #134. As it happens, the difference between the incorrect and correct sigmoidPrime implementations is an application of sigmoid. The difference between hidden_sum and activated_hidden is also an application of sigmoid. So that means, self.sigmoidPrime(self.activated_hidden) (using the incorrect definition of sigmoidPrime) is equivalent to self.sigmoidPrime(self.hidden_sum) (using the correct definition for sigmoidPrime). This means that the implementation of NeuralNetwork would break if sigmoid were swapped with just about any other activation function, but it works with sigmoid, seemingly by accident.

Caveat:
I do not know if this is related specific to my local setup or a more general issue.

cd into root directory of DS-UNIT-4-SPRINT-2.

Run command:
conda env create -f environment.yml -n DS-4.2

Returns ResolvePackageNotFound: errors for particular packages.

Resolve By:
Move the packages in the error inside yml from dependencies to pip

My Understanding/Guesstimation:
These are packages that are required but are not installed?
Thus adding to pip in yml has them installed instead of fetched from local?

Additional Errors attempting to create conda env:

CondaValueError: prefix already exists:

Resolve By:
add --force to conda env create command

NOTE: There are packages in the requirements.txt file that are not in dependencies or pip inside yml. So you still need to run pip install on the requirements.txt inside the new conda env.

If after running pip install -r requirements.txt you get

ERROR: Could not install packages due to an EnvironmentError:
then add --user to the command:
pip install -r requirements.txt --user

This yml/requirements.txt installs regular tensorflow and not tensorflow-gpu but thats prob intentional.

CSV link from Google sheets is broken. :(

In the .ipynb for assignment 3, it says "Data Science Unit 4 Sprint 2 Assignmnet 3" Assignment is misspelled :)

In module1-Intro-to-Neural-Networks/LS_DS_421_Intro_to_NN_Assignment.ipynb, The final cell defining the perceptron class there needs to be an additional indent on line 6: 'self.niter = niter'.

Everything within the 'fit' function needs to be indented.

The docstring at the begining of the predict function needs indentation