DM_08_04

Import packages

We'll create a hidden Markov model to examine the state-shifting in the dataset.



In [3]:

    
% matplotlib inline

import pylab
import numpy as np
import pandas as pd

from hmmlearn.hmm import GaussianHMM

Import data

Read CSV file into "df."



In [4]:

    
df = pd.read_csv("speed.csv", sep = ",")
df.head(5)

Drop the row number and "corr" so we can focus on the influence of "prev" and "Pacc" on "rt." Also define "prev" as a factor.



In [5]:

    
x = df.drop(["row", "corr"], axis = 1)
x["prev"] = pd.factorize(x["prev"])[0]

Create model

Make an HMM with 2 states. (The choice of 2 is based on theory.)



In [6]:

    
model = GaussianHMM(n_components=2, n_iter=10000, random_state=1).fit(x)
model.monitor_









    Out[6]:





ConvergenceMonitor(history=[-116.44940310313338, -116.4467051828148], iter=8,
          n_iter=10000, tol=0.01, verbose=False)

Predict the hidden state for each record and get count of predicted states.



In [7]:

    
states = model.predict(x)
pd.Series(states).value_counts()









    Out[7]:





0    243
1    196
dtype: int64

Get the mean reaction time (rt) for each of the two states.



In [8]:

    
model.means_[:, 0]









    Out[8]:





array([ 6.40098754,  5.56434105])

Visualize results

Make a graph to show the change of states.



In [9]:

    
fig = pylab.figure(figsize=(20, 1))
ax = fig.add_subplot(111)
ax.grid(True)
ax.set_xlabel("Record number")
ax.set_ylabel("State")
ax.plot(states)









    Out[9]:





[<matplotlib.lines.Line2D at 0x116735b50>]



In [ ]:

	row	rt	corr	prev
0	1	6.456770	cor	inc
1	2	5.602119	cor	cor
2	3	6.253829	inc	cor
3	4	5.451038	inc	inc
4	5	5.872118	inc	inc