Random Forest

random_forest.py

import pandas as pd
import numpy as np
from sklearn.ensemble import RandomForestClassifier
import sklearn.metrics as skm
import pylab as pl

#Read in the column names for the dataset
feat = pd.read_csv('features.txt', delim_whitespace=True, header=None, index_col=False)

#Name the columns to isolate all dataset column names in one column
feat.columns=['column 1', 'column 2']

#Clean up column names
feat['column 2'] = feat['column 2'].str.replace('-', '')
feat['column 2'] = feat['column 2'].str.replace('(', '')
feat['column 2'] = feat['column 2'].str.replace(')', '')
feat['column 2'] = feat['column 2'].str.replace(',', '')
feat['column 2'] = feat['column 2'].str.replace('BodyBody', '')
feat['column 2'] = feat['column 2'].str.replace('Body', '')
feat['column 2'] = feat['column 2'].str.replace('Mag', '')
feat['column 2'] = feat['column 2'].str.replace('mean', 'Mean')
feat['column 2'] = feat['column 2'].str.replace('std', 'STD')

#Read in raw data
#The 'y_train' and 'subject_train' files are in ASCII character format
X_train = pd.read_csv('train/X_train.txt', delim_whitespace=True, header=None, index_col=False)
y_train = pd.read_csv('train/y_train.txt', header=None, index_col=False)
subjects = pd.read_csv('train/subject_train.txt', header=None, index_col=False)

#Isolate clean column names from 'feat' DataFrame
features=feat['column 2']

## Assign column names to each raw data file
X_train.columns = features
y_train.columns = ['Activity']
subjects.columns = ['Subject']

#Remove duplicate columns from X_train file
X_train = X_train.loc[:,~X_train.columns.duplicated()]

#merge all dataframes together
data = pd.merge(y_train, X_train, left_index=True, right_index=True)
data = pd.merge(data, subjects, left_index=True, right_index=True)

#make 'Activity' a categorical variable
data['Activity'] = pd.Categorical(data['Activity']).codes

#Separate testing and training data
train = data.query('Subject >= 27')
test = data.query('Subject <= 6')
validation = data.query('(Subject >= 21) & (Subject < 27)')

#Fit random forest to training set
train_target = train['Activity']
#Remove unnecessary columns
train_data = train.ix[:,1:-2]
rfc = RandomForestClassifier(n_estimators=50, oob_score=True)
rfc.fit(train_data, train_target)

print('Out of the box score: ', rfc.oob_score_)

#Most important features
importances = rfc.feature_importances_
indices = np.argsort(importances)[::-1]
print('Top 10 Features:')
for i in range(10):
	print("%d. feature %d (%f)" % (i + 1, indices[i], importances[indices[i]]))
	#Not sure how to get the column names to print, rather than their index

#Define validation set and make accuracy score predictions
val_target = validation['Activity']
#Remove unnecessary columns
val_data = validation.ix[:,1:-2]
#Calculate accuracy score
val_pred = rfc.predict(val_data)

#Define test set and make accuracy score predictions
test_target = test['Activity']
#Remove unnecessary columns
test_data = test.ix[:,1:-2]
#Calculate accuracy score
test_pred = rfc.predict(test_data)

#Print accuracy scores
print("Validation set Mean Accuracy Score: %f" % (rfc.score(val_data, val_target)))
print("Test set Mean Accuracy Score: %f" % (rfc.score(test_data, test_target)))

#Confusion matrix
test_cm = skm.confusion_matrix(test_target, test_pred)
pl.matshow(test_cm)
pl.title('Test data confusion matrix')
pl.colorbar()
pl.show()

#Print accuracy, precision, recall, and f1 scores for test set
print("Accuracy = %f" %(skm.accuracy_score(test_target, test_pred)))
print("Precision = %f" %(skm.precision_score(test_target, test_pred, average='weighted')))
print("Recall = %f" %(skm.recall_score(test_target, test_pred, average='weighted')))
print("F1 score = %f" %(skm.f1_score(test_target, test_pred, average='weighted')))

###### Note: I had to work off of another students project to figure this assignmant out fully.