Python监督机器学习

Question

I am trying to understand how to use scikit for supervised machine learning so I've made up some data belonging to two different sets: set A and set B. I have 18 elements in set A and 18 elements in set B. Each of the elements have three variables. See below:

#SetA
Variable1A = [ 3,4,4,5,4,5,5,6,7,7,5,4,5,6,4,9,3,4]
Variable2A = [ 5,4,4,3,4,5,4,5,4,3,4,5,3,4,3,4,4,3]
Variable3A = [ 7,8,4,5,6,7,3,3,3,4,4,9,7,6,8,6,7,8]


#SetB
Variable1B = [ 7,8,11,12,7,9,8,7,8,11,15,9,7,6,9,9,7,11]
Variable2B = [ 1,2,3,3,4,2,4,1,0,1,2,1,3,4,3,1,2,3]
Variable3B = [ 12,18,14,15,16,17,13,13,13,14,14,19,17,16,18,16,17,18]

How would I use scikit to use supervised machine learning so that when I introduce a new setA and setB data it can try to identify which of the new data belongs to either setA or setB.

Apologies for the data sets are small and 'made up'. I just want to apply the same method using scikit on other data sets.

Answer 1

Your question is quite broad, so this is just a brief outline. Instead of formatting your data that way, you want to put the two sets together in one list/array, with another column to represent which set each row belongs to. Something like this:

data = [
    [3, 5, 7, 0]
    [4, 4, 8, 0],  # these rows have 0 as the last element to represent group A
    ...
    [7, 1, 12, 1],
    [8, 2, 18, 1], # these have 1 as the last element to represent group A
    ...
]

An alternative is to put only the first three columns in data and call it X , and then have a separate array y containing just [0, 0, 0, ..., 1, 1, 1, ...] (indicating group membership of each row). What you want to avoid is having the information about which group a point is in be stored in the name of the variables; you instead want to have the "set A or set B" information stored in the values of variables (as here it's stored in the values in the last column of data , or in y ),

Whatever you do, you'll almost certainly want to use numpy arrays or pandas data structures to hold your data, rather than lists.

There are numerous tutorials and examples available for how to use scikit-learn, as well as sample data sets that may be more useful than the one you made up. "Supervised machine learning" is a broad term incorporating many different approaches to the task of deciding which group a data point is in, so you'll have to play around and try out different classification algorithms. All of this info can be found by googling and/or browsing through the scikit documentation.

Answer 2

I think this is a good question and no worries if you get the feeling it was not clear enough. Supervised learning can be used to classify an instance (data row) into several categories (or in your case just 2 sets). What you are missing in the above example is a variable that says in which set 1 row belongs to.

import numpy as np # numpy will help us to concatenate the columns into a 2-dimensional array
# so instead of hiving 3 separate arrays, we have 1 array with 3 columns and 18 rows 

Variable1A = [ 3,4,4,5,4,5,5,6,7,7,5,4,5,6,4,9,3,4]
Variable2A = [ 5,4,4,3,4,5,4,5,4,3,4,5,3,4,3,4,4,3]
Variable3A = [ 7,8,4,5,6,7,3,3,3,4,4,9,7,6,8,6,7,8]

#our target variable for A

target_variable_A=[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]

Variable1B = [ 7,8,11,12,7,9,8,7,8,11,15,9,7,6,9,9,7,11]
Variable2B = [ 1,2,3,3,4,2,4,1,0,1,2,1,3,4,3,1,2,3]
Variable3B = [ 12,18,14,15,16,17,13,13,13,14,14,19,17,16,18,16,17,18]

# target variable for B
target_variable_B=[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]

#lets create a dataset C with only 4 rows that we need to predict if belongs to "1" which is data set A or "0" which is  dataset B

Variable1C = [ 7,4,4,12]
Variable2C = [ 1,4,4,3]
Variable3C = [ 12,8,4,15]

#make the objects 2-dimenionsal arrays (so 1 array with X rows and 3 columns-variables)
Dataset_A=np.column_stack((Variable1A,Variable2A,Variable3A))
Dataset_B=np.column_stack((Variable1B,Variable2B,Variable3B))
Dataset_C=np.column_stack((Variable1C,Variable2C,Variable3C))

print(" dataset A rows ", Dataset_A.shape[0]," dataset A columns ", Dataset_A.shape[1] )
print(" dataset B rows ", Dataset_B.shape[0]," dataset B columns ", Dataset_B.shape[1] )
print(" dataset C rows ", Dataset_C.shape[0]," dataset C columns ", Dataset_C.shape[1] )

##########Prints ##########
#(' dataset A rows ', 18L, ' dataset A columns ', 3L)
#(' dataset B rows ', 18L, ' dataset B columns ', 3L)
#(' dataset C rows ', 4L, ' dataset C columns ', 3L)

# since now we have an identification that tells us if it belongs to A or B (e.g. 1 or 0) we can append the new sets together
Dataset_AB=np.concatenate((Dataset_A,Dataset_B),axis=0) # this creates a set with 36 rows and 3 columns
target_variable_AB=np.concatenate((target_variable_A,target_variable_B),axis=0)

print(" dataset AB rows ", Dataset_AB.shape[0]," dataset Ab columns ", Dataset_AB.shape[1] )
print(" target Variable rows ", target_variable_AB.shape[0])

##########Prints ##########
#(' dataset AB rows ', 36L, ' dataset Ab columns ', 3L)
#(' target Variable rows ', 36L)

#now we will select the most common supervised scikit model - Logistic Regression
from sklearn.linear_model import LogisticRegression
model=LogisticRegression() # we create an instance of the model

model.fit(Dataset_AB,target_variable_AB) # the model learns to distinguish between A and B (1 or 0)

#now we make predictions for the new dataset C

predictions_for_C=model.predict(Dataset_C)
print(predictions_for_C)
# this will print
#[0 1 1 0]
# so first case belongs to set A , second to B, third to B and fourth to A

Answer 3

Supervised learning means that the data you are providing for training the model is labelled that is the outcome of each sample used for training is known before hand.

In the problem you have provided there are basically 2 sets : set A and set B so you will have to use a binary classifier like Logistic Regression model .

First label elements of set A and B as either 1 or 0 vice versa based on which set they belong to , that is say if element e belongs to set A marks it as 1 else mark it as 0.

Then import the Logistic Regression classifier from scikitlearn in python.

Next thing is that merge both the sets like set A followed by set B or vice versa and in the same order merge the labels you have already provided .

You can use either pandas or numpy for stacking these sets up and preparing the labelled dataset.

Now you have a good well labelled dataset .

You can now call the fit function from the Logistic Regression Classifier with the dataset(containing set A and set B elements) and the label set.

After that call the predict function with the data you want to test it with you will get the predicted class that is either 0 or 1.

If you want the sets you can use a dictionary to map the keys as 1 and 0 with values 'set A' and 'set B' . So that you can get the sets from that.

import pandas as pd
import numpy as np 
from sklearn.linear_model import LogisticRegression as lr

#set A

firstA=[3,4,4,5,4,5,5,6,7,7,5,4,5,6,4,9,3,4]
secondA=[5,4,4,3,4,5,4,5,4,3,4,5,3,4,3,4,4,3]
thirdA=[7,8,4,5,6,7,3,3,3,4,4,9,7,6,8,6,7,8]

#set B

firstB=[7,8,11,12,7,9,8,7,8,11,15,9,7,6,9,9,7,11]
secondB=[1,2,3,3,4,2,4,1,0,1,2,1,3,4,3,1,2,3]
thirdB=[12,18,14,15,16,17,13,13,13,14,14,19,17,16,18,16,17,18]

#stacking up and building the dataset

Aset=[firstA,secondA,thirdA]
Bset=[firstB,secondB,thirdB]
totalset=[Aset,Bset]


data=pd.DataFrame(columns["0","1","2","3","4","5","6",
"7","8","9","10","11","12","13","14","15","16","17"])
c=0
for i in range(0,2):
    for j in range(0,3):
        data.loc[c]=totalset[i][j]
        c=c+1 
label=np.array([0,0,0,1,1,1])
df2=pd.DataFrame(columns=["0","1","2","3","4","5"])
df2=label


#Training and testing the model

model=lr()
model.fit(df,df2)
k=model.predict([[17,18,14,15,16,17,13,
13,13,41,14,19,17,16,18,16,17,28]])

#mapping(chosen set A element's with label 0 and set B with 1)

dic={0:"set A",1:"set B"}
print(dic[int(k)])