4 Using_Sklearn

##We should start by loading any packages we will need, as well as any data.

## We can import a full package, but I'm just going to import some:

## sklearn things we need:
#from sklearn.neighbors import KNeighborsClassifier
from sklearn import neighbors
from sklearn.neighbors import *
from sklearn.model_selection import train_test_split
from sklearn import datasets

# numpy and friends
import numpy as np
import matplotlib.pyplot as plt
from scipy import stats
%matplotlib inline

## We don't need all of this, but it's good to get anything I need in the very first block.

## Let's start with the handwritten numbers that we've seen a lot of:

digits = datasets.load_digits()

X = digits['data']   # this is the data with each 8x8 image "flattened" into a length-64 vector.
Y = digits['target'] # these are the labels (0-9). 

## Let's look at one number:
## I've chosen randomly, but set n to any number under 1000


n = 653

plt.imshow(digits['images'][n], cmap='Greys_r')
plt.title('This is a %d' % digits['target'][n])

print("The X values are a 64 digit string:", X[n])
print("and the y is the value of the number", Y[n])

## The first step is to break the data into the training set, and the test set.  
## Luckily, sklearn has you covered:

X_train, X_test, Y_train, Y_test = train_test_split(X,Y, test_size=0.2)

## test_size tells the algrithm that we want to reserve 20% for the test (or validation) set.  Try it!  
## Now redo it with a test set at 30%

## Solutions

## Lets do it!
from sklearn.neighbors import *

## Run this to learn about the function:
?KNeighborsClassifier

## FYI: Minkowski distance means the Lp norm, so we can set p = 2 for Euclidean distance

## For now, I'm going to set K = 3, and use the Euclidean or L2 norm.  

#Pick a name for your model -> I'm going to use a random word: cupcake

#step 1: let python know what algorithm I'm using:
cupcake =neighbors.KNeighborsClassifier(n_neighbors=15, p =2 )
#step 2: Train the model on ONLY THE TRAINING DATA (fit!)
cupcake.fit(X_train,Y_train)


##Now we have a well trained model!  But is it a good model?
## I'm going to look at the errors in BOTH my training and test sets.  
## We do this by predicting what Y will be for each set, than actually comparing

Y_pred_train = cupcake.predict(X_train) 
Y_pred = cupcake.predict(X_test)

##record errors
error_train=list(Y_train == Y_pred_train).count(False)/len(Y_train)
error_test=list(Y_test == Y_pred).count(False)/len(Y_test)

##Look it over:
print("The training error was: ", error_train)
print("The testing error was: ", error_test)