# Naming A Baby Part 1

## Introduction¶

Last year my wife and I had a child, Garen Jay Marrs. Finding a name for him was an interesting task due to my wife having difficulties in picking names. Our first pet together, a cat, did not have a name for around 2 months.

I spent quite a bit of time researching different apps available, but they did not seem to do what I wanted. Generally the apps only offered going through a list of names and liking or disliking them. As a data scientist, I was hoping to find an app that would suggest names as you provided feedback on which names you like and dislike. None of the apps that I reviewed had this capability so I decided to create my own.

This post is the starting point of a short blog series that describes how I created my ideal baby naming application. It will cover both high and low level details of how it works and my reasoning behind the functionality. At the end of this blog series, the application will be freely available.

## What Is In A Name?¶

When choosing a name, most of us try to sound out the name and just create a list of what we like. However, when you look at it from an analytical eye, you see patterns in the names. Some features that my wife and I found important when choosing a name for our son included:

• The letter it starts with
• 2 - 3 letters that the name starts with
• 2 - 3 letters that the name ends with
• What it sounds like when pronounced
• Percentage of the population that have the name as male or female

## Dataset¶

The dataset that I used is from the Social Security Administration. They provide various statistics (percent male and female) over a period of time for names. This was a great set of names consisting of around 10,000 names. Unfortunately, it does not include a good mix of various ethnicities; Arabic, Chinese and Indian to name a few. For our case this was not an issue.

https://www.ssa.gov/OACT/babynames/

# Anomaly Detection: Matrix Profile Discords

Matrix profiles are used to annotate a time series in a way that makes data mining tasks within the time series simple. I will not explain what a matrix profile is within this post. Feel free to visit this page to understand more about them:

https://www.cs.ucr.edu/~eamonn/MatrixProfile.html

During the KDD 2017 conference, I had the pleasure to be introduced to the concept of the matrix profiles. Unfortunately, at the time all of the code was implemented in MatLab. I dabbled a little bit with converting the code to Python and never completed it. Last month, December 2018, someone (a Target employee) created a github repository with working algorithms. Here is the github repository:

https://github.com/target/matrixprofile-ts

My main interest of matrix profiles was the usefulness in anomaly detection. This blog post is going to demonstrate how to use the Python module to detect anomalies within a NAB dataset. Specifically, I am working with the NYC Taxi dataset.

### Data Overview¶

The data consists of the number of taxi passengers from 2014-07-01 to 2015-01-31. There are 5 known anomalies during these periods:

• NYC Marathon - 2014-11-02
• Thanksgiving - 2014-11-27
• Christmas - 2014-12-25
• New Years - 2015-01-01
• Snow Blizzard - 2015-01-26 and 2015-01-27

I will see how close the anomaly detection is using matrix profiles.

In [2]:
from matrixprofile import *
from matrixprofile.discords import discords

In [3]:
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt

%matplotlib inline


In [4]:
df = pd.read_csv('/home/tyler/data/nab/realKnownCause/realKnownCause/nyc_taxi.csv')

In [5]:
df['timestamp'] = pd.to_datetime(df['timestamp'])

In [6]:
df = df.set_index('timestamp').sort_index()

In [7]:
df.head()

Out[7]:
value
timestamp
2014-07-01 00:00:00 10844
2014-07-01 00:30:00 8127
2014-07-01 01:00:00 6210
2014-07-01 01:30:00 4656
2014-07-01 02:00:00 3820

### Resample Hourly¶

Originally the dataset is within 30 minute increments.

In [8]:
df = df.resample('1H').sum()

In [9]:
df.head()

Out[9]:
value
timestamp
2014-07-01 00:00:00 18971
2014-07-01 01:00:00 10866
2014-07-01 02:00:00 6693
2014-07-01 03:00:00 4433
2014-07-01 04:00:00 4379
In [10]:
a = df.values.squeeze()

# subsequence length to compute the matrix profile
# since we have hourly measurements and want to find daily events,
# we will create a length of 24 - number of hours in a day
m = 24
profile = matrixProfile.stomp(a,m)

In [11]:
df['profile'] = np.append(profile[0],np.zeros(m-1)+np.nan)


### Plot Matrix Profile¶

Below is a plot of the hourly data and the matrix profile. Visually, you can see both motifs and discords. We are interested in finding the discords which are high peaks in the plot. A couple of periods jump out that seem close to Thanksgiving and the snow storm.

In [12]:
#Plot the signal data
fig, (ax1, ax2) = plt.subplots(2,1,sharex=True,figsize=(15,10))
df['value'].plot(ax=ax1, title='Raw Data')

#Plot the Matrix Profile
df['profile'].plot(ax=ax2, c='r', title='Matrix Profile')
plt.show()

In [13]:
# exclude up to a day on the left and right side
ex_zone = 24

# we look for the 5 events specified in the data explaination
anoms = discords(df['profile'], ex_zone, k=5)

In [14]:
df.iloc[anoms]

Out[14]:
value profile
timestamp
2015-01-27 09:00:00 3874 3.275818
2014-11-02 00:00:00 48219 2.468237
2015-01-25 20:00:00 29503 2.094151
2014-12-31 23:00:00 35978 1.803195
2014-12-24 00:00:00 20646 1.581157

### Conclusions¶

Using the matrix profile to identify discords within the NYC taxi dataset seems fruitful. All of the anomalies mentioned in the dataset overview were found. At first it can be a little tricky to think about what subsequence length to use, but once you understand your problem it becomes clear. I will be using this module more frequently when I need to identify anomalies in the future.

# Deriving Markov Transition Matrices

## Introduction¶

I had a need at work to understand the flow of a system. Knowing that a transition matrix would be perfect to understand state transitions, I researched available libraries within the Python ecosystem. Unfortunately, I could not find a suitable solution.

My desires in a library included:

• It must capture states as strings or numbers for easy analysis.
• Store both frequencies and probabilities of state transitions.
• Allow simple filtering of frequencies and probabilities.
• Output the derived matrix as a panadas dataframe or numpy matrix.
• Plot the matrix as a Digraph.

This blog post shows my method of deriving a Markov transition matrix that covers all of those desires. This blog post DOES NOT discuss what a Markov transition matrix is in depth. You should read more about them on your own.

## Code¶

The code block below contains two important classes; Transition and TransitionMatrix. Transition is a class that keeps track of the current state and next state along with the frequency and probability. The TransitionMatrix stores all state transitions and provides methods for plotting and merging.

Note The graphviz and pandas modules are required to run this code.

In [1]:
"""
This module is used to compute Markov Chains.
"""
import itertools
import copy

import pandas as pd

from graphviz import Digraph
from graphviz import Source

class Transition(object):
"""
This class hold meta-data about a transition between a Markov stochastic process.
Essentially it is used to keep track of frequencies and to compute probability of the
transition.
"""
def __init__(self, current_state, next_state):
"""
Creates a transition object given the current_state and next_state as strings.

Parameters
----------
current_state : str
The current state of this transition.
next_state : str
The next state of this transition.
"""
self.current_state = current_state
self.next_state = next_state
self.prob = 0
self.freq = 0

def increment_freq(self):
self.freq += 1

def increment_freq_by(self, x):
self.freq += x

def compute_prob(self, current_state_freqs):
try:
self.prob = self.freq / current_state_freqs
except ZeroDivisionError:
pass

def __hash__(self):
return hash((self.current_state, self.next_state))

def __eq__(self, other):
if not isinstance(other, Transition):
return False

return self.current_state == other.current_state \
and self.next_state == other.next_state

def __str__(self):
return '{} -> {}'.format(self.current_state, self.next_state)

def __copy__(self):
tmp = Transition(self.current_state, self.next_state)
tmp.increment_freq_by(self.freq)

return tmp

class TransitionMatrix(object):
"""
The transition matrix is an object that is used to derive the Stochastic matrix
of a Markov chain. Internally, everything is storing in a dictionary of transitions.
This class also provides methods to obtain dict of dict, pandas dataframe, dotlang or
a graph representation of the matrix.
"""
def __init__(self, states, valid_states=None):
"""
This kicks off the computation of the matrix given states and optionally valid states.

Parameters
----------
states : :obj:list of :obj:obj
The states to derive the matrix from. These states can be either int or str.
valid_states : :obj:list of :obj:obj, optional
Optionally include valid states that are used to derive all valid transitions.
"""
self.transitions = {}
self.state_frequencies = {}

if valid_states:
self.__generate_transitions(valid_states)
else:
self.__generate_transitions(states)

self.__compute_matrix(states)

def __generate_transitions(self, states):
"""
Internal method that generates all permutations of transitions.
"""
tmp = list(itertools.permutations(states, 2))

current = None
for t in tmp:
if t[0] not in self.state_frequencies:
self.state_frequencies[t[0]] = 0

if current != t[0]:
transition = Transition(t[0], t[0])
self.transitions[transition] = transition
current = t[0]

transition = Transition(t[0], t[1])
self.transitions[transition] = transition

def __compute_matrix(self, states):
"""
Internal method that computes frequencies and probability of the matrix.
"""
for i, j in zip(states, states[1:]):
self.transitions[Transition(i, j)].increment_freq()
self.state_frequencies[i] += 1

self.__compute_prob()

def __compute_prob(self):
"""
Internal method that computes the probability of all transitions.
"""
for transition in self.transitions:
self.transitions[transition].compute_prob(
self.state_frequencies[self.transitions[transition].current_state]
)

def __recompute_from_transitions(self):
"""
Internal method that is primarily used when filtering and/or combining transition
matrices together. It is used to derive current state frequencies from transitions.
This can then be used to compute the correct probabilities.
"""
self.state_frequencies = {}
for t in self.transitions:
c = self.transitions[t].current_state
freq = self.transitions[t].freq

if c not in self.state_frequencies:
self.state_frequencies[c] = 0

if freq > 0:
self.state_frequencies[c] += freq

self.__compute_prob()

def as_dict(self, prob=False, include_all=False):
"""
Converts the transition matrix into a dict of dicts.
"""
tmp = {}
for t in self.transitions:
c = self.transitions[t].current_state
n = self.transitions[t].next_state

val = self.transitions[t].freq
if prob:
val = self.transitions[t].prob

if val <= 0 and not include_all:
continue

if c not in tmp:
tmp[c] = {}

tmp[c][n] = val

return tmp

def as_dataframe(self, prob=False, include_all=False):
return pd.DataFrame(self.as_dict(prob=prob, include_all=include_all))

def as_dotlang(self):
data = []
for t in self.transitions:
c = self.transitions[t].current_state
n = self.transitions[t].next_state
prob = self.transitions[t].prob

if prob > 0:
t = '   "{}" -> "{}" [label = "{:2.4f}"]'.format(c, n, prob)
data.append(t)

return "digraph G {\n%s\n}" % ('\n'.join(data))

def as_graph(self):
g = Digraph(format='svg', engine='dot')

for t in self.transitions:
c = self.transitions[t].current_state
n = self.transitions[t].next_state
prob = self.transitions[t].prob

if prob > 0:
g.edge(c, n, label=" {:2.4f} ".format(prob))

return g


## Example¶

In this example, I will generate a handful of states that occur linearly. The TransitionMatrix expects a list of values as the states or a string. If a string is passed, it treats each character in the string as a state.

In [2]:
import numpy as np

In [3]:
states_a = [np.random.choice(['A', 'B', 'C'], p=[0.2, 0.5, 0.3]) for i in range(1000)]


Passing the states into the matrix derives the frequencies and probabilities.

In [4]:
transition_matrix = TransitionMatrix(states_a)


### DataFrame¶

To obtain the matrix as a pandas dataframe we call the as_dataframe() method. By default we show the frequencies, however we can also get the probabilities.

In [5]:
transition_matrix.as_dataframe()

Out[5]:
A B C
A 41 108 59
B 103 238 146
C 63 142 99
In [6]:
transition_matrix.as_dataframe(prob=True)

Out[6]:
A B C
A 0.198068 0.221311 0.194079
B 0.497585 0.487705 0.480263
C 0.304348 0.290984 0.325658

### Visualize¶

We are able to visualize the transition matrix as a Digraph by calling the as_graph() method. If we wanted to export a dotfile, we can call the as_dotlang() method to obtain the dotlang string.

In [7]:
transition_matrix.as_graph()

Out[7]:

### Filtering¶

Filtering is easily accomplished with a pandas dataframe.

In [8]:
df = transition_matrix.as_dataframe(prob=True)

In [9]:
df[df > 0.2]
`
Out[9]:
A B C
A NaN 0.221311 NaN
B 0.497585 0.487705 0.480263
C 0.304348 0.290984 0.325658

You can see that we filtered on a probability of 0.2 creating a NaN probability between C and A.