In [1]:
import pandas as pd
from mlxtend.frequent_patterns import apriori
from mlxtend.frequent_patterns import association_rules
import numpy as np
import matplotlib.pyplot as plt
pd.set_option('display.max_columns',100)
#pd.set_option('display.width',2000)
pd.set_option('display.max_colwidth', -1)
In [2]:
#lets have a look at the data file
import csv
with open('./groceries.csv', newline='') as csvfile:
reader = csv.reader(csvfile, delimiter=',', quotechar='|')
for row_number, row in enumerate(reader):
print("[{}] {}".format(row_number,', '.join(row)))
if (row_number>10):
break
In [3]:
#load dataset
df = pd.read_table('./groceries.csv',header=None)
#create binary matrix to be used as input to apriori
df1= df.iloc[:,0].str.get_dummies(sep=',')
df1.head()
Out[3]:
In [4]:
#count number of transactions in dataset
len(df1.index)
Out[4]:
In [5]:
#how many transactions contain beef
df1['beef'].sum()
Out[5]:
In [6]:
#count number of transactions per product
df1.sum().nlargest(10)
Out[6]:
In [7]:
#plot products with most transactions
df1.sum().nlargest(10).plot.bar(title='Top-10 products with their cumulative sales')
Out[7]:
In [19]:
#plot products with most transactions
df1.sum().nsmallest(10).plot.bar(title='Bottom-10 products with their cumulative sales')
Out[19]:
In [9]:
#row sum computes number of products in a basket
dfcounts=df1.sum(axis=1)
plt.xlabel('Number of items in a transaction')
plt.ylabel('Number of Transactions')
plt.hist(dfcounts,bins=range(1,32),rwidth=0.8)
plt.show()
In [10]:
# Build up the frequent items
frequent_itemsets = apriori(df1, min_support=0.002, use_colnames=True)
frequent_itemsets
Out[10]:
In [11]:
#plot itemsets with exactly 3 items with highest support values
frequent_itemsets[frequent_itemsets.itemsets.str.len()==3].sort_values(by=['support'],ascending=False).head()
Out[11]:
In [12]:
# Create the rules
rules = association_rules(frequent_itemsets)
rules[['antecedents','consequents','support','confidence','lift']]
Out[12]:
In [13]:
#list rules with high lift and confidence values
rules[ (rules['lift'] >= 4) &
(rules['confidence'] >= 0.8) ] [['antecedents','consequents','support','confidence','lift']]
Out[13]:
In [14]:
def draw(rules):
import networkx as nx
G = nx.DiGraph()
for i,row in rules.iterrows():
for c in row['consequents']:
G.add_nodes_from([c])
for a in row['antecedents']:
G.add_nodes_from([a])
G.add_edge(a,c,color='black',weight=row['confidence'])
color_map=[]
for node in G:
#color blue nodes that participate as a consequent to a rule
if (rules['consequents'].str.contains(node, regex=False).any()):
color_map.append('blue')
else:
color_map.append('red')
edges = G.edges()
colors = [G[u][v]['color'] for u,v in edges]
weights = [G[u][v]['weight'] for u,v in edges]
min_weight=min(weights)-0.01
max_weight=max(weights)
weights = [5*(w-min_weight)/(max_weight-min_weight) for w in weights]
pos = nx.spring_layout(G, k=10, scale=1)
nx.draw(G, pos,edges=edges,edge_color=colors,node_color=color_map,width=weights,font_size=8,with_labels=False)
for p in pos: # raise text positions
pos[p][1] += 0.18
nx.draw_networkx_labels(G, pos)
plt.show()
In [15]:
draw(rules)
In [16]:
#list rules that contain 'citrus fruit' in the LHS
rules[rules['antecedents'].apply(str).str.contains('citrus fruit')][['antecedents','consequents','support','confidence','lift']]
Out[16]:
In [17]:
rules.plot.scatter(x='support',y='confidence',c='lift', colormap='viridis')
Out[17]:
In [18]:
rules[ (rules['lift'] >= 4.4)][['antecedents','consequents','support','confidence','lift']]
Out[18]:
In [ ]: