-
Notifications
You must be signed in to change notification settings - Fork 0
/
star-trek.py
167 lines (118 loc) · 4.51 KB
/
star-trek.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
# Write-up: https://ology.github.io/2020/03/03/predicting-star-trek-characters-with-naive-bayes/
# LOAD FILES INTO A DATAFRAME
import pandas as pd
import nltk.data
def file_to_df(path, name):
fh = open(path + '/' + name + '.txt')
data = fh.read()
fh.close()
tokenizer = nltk.data.load('nltk_data/tokenizers/punkt/english.pickle')
return pd.DataFrame(
zip(
tokenizer.tokenize(data),
([name] * len(tokenizer.tokenize(data)))
),
columns=['text','person']
)
path = 'Kirk-Spock-McCoy'
mccoy = file_to_df(path, 'mccoy')
spock = file_to_df(path, 'spock')
kirk = file_to_df(path, 'kirk')
result = pd.concat([mccoy, spock, kirk])
# SHOW COUNTS
import matplotlib.pyplot as plt
result.groupby('person').text.count().plot.bar(ylim=0)
plt.show()
# GET THE TRAIN/TEST DATA
X = result.text
y = result.person
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)
# LEARN THE VOCABULARY
from sklearn.feature_extraction.text import CountVectorizer
vect = CountVectorizer() #stop_words='english' => 0.6598099205832574
X_train_dtm = vect.fit_transform(X_train)
X_test_dtm = vect.transform(X_test)
X_dtm = vect.transform(X)
# TF-IDF SCALE THE VOCABULARY
from sklearn.feature_extraction.text import TfidfTransformer
#tfidf_transformer = TfidfTransformer()
#X_train_tfidf = tfidf_transformer.fit_transform(X_train_dtm)
#X_test_tfidf = tfidf_transformer.fit_transform(X_test_dtm)
# TRAIN A CLASSIFIER
from sklearn.naive_bayes import MultinomialNB
#clf = MultinomialNB().fit(X_train_tfidf, y_train)
#y_pred = clf.predict(X_test_tfidf) # 0.628563989063924
clf = MultinomialNB().fit(X_train_dtm, y_train)
y_pred = clf.predict(X_test_dtm) # 0.6656685327431324
# EVALUATE THE MODEL ACCURACY
from sklearn import metrics
metrics.accuracy_score(y_test, y_pred)
#metrics.confusion_matrix(y_test, y_pred)
# PREDICT ARBITRARY PHRASES
docs = [ "Captian's log, Stardate...", 'That is highly illogical.', "He's dead Jim." ]
X_new_counts = vect.transform(docs)
predicted = clf.predict(X_new_counts)
for doc, who in zip(docs, predicted):
print('%r => %s' % (doc, who))
# MOST PREDICTIVE TOKENS
vect = CountVectorizer(stop_words='english')
X_train_dtm = vect.fit_transform(X_train)
X_test_dtm = vect.transform(X_test)
tfidf_transformer = TfidfTransformer()
X_train_tfidf = tfidf_transformer.fit_transform(X_train_dtm)
X_test_tfidf = tfidf_transformer.fit_transform(X_test_dtm)
clf = MultinomialNB().fit(X_train_tfidf, y_train)
X_train_tokens = vect.get_feature_names()
kirk_token_count = clf.feature_count_[0, :]
mccoy_token_count = clf.feature_count_[1, :]
spock_token_count = clf.feature_count_[2, :]
tokens = pd.DataFrame(
{
'token': X_train_tokens,
'mccoy': mccoy_token_count,
'spock': spock_token_count,
'kirk': kirk_token_count
}
).set_index('token')
tokens.head()
tokens['kirk'] = tokens.kirk + 1
tokens['mccoy'] = tokens.mccoy + 1
tokens['spock'] = tokens.spock + 1
tokens['kirk'] = tokens.kirk / clf.class_count_[0]
tokens['mccoy'] = tokens.mccoy / clf.class_count_[1]
tokens['spock'] = tokens.spock / clf.class_count_[2]
tokens.sort_values('kirk', ascending=False).head(10)
tokens.sort_values('mccoy', ascending=False).head(10)
tokens.sort_values('spock', ascending=False).head(10)
# Some more phrases!
def who_said(docs):
X_new_counts = vect.transform(docs)
predicted = clf.predict(X_new_counts)
for doc, who in zip(docs, predicted):
print('%r => %s' % (doc, who))
docs = [ 'take me to the doctor, captain', 'jim, where is spock?' ]
who_said(docs)
# 'take me to the doctor, captain' => spock
# 'jim, where is spock?' => mccoy
docs = [ 'our father who art in heaven' ]
who_said(docs)
# 'our father who art in heaven' => kirk
# SVM
#from sklearn.linear_model import SGDClassifier
#clf = SGDClassifier(loss='hinge', penalty='l2', alpha=1e-3, n_iter=5, random_state=42)
#clf.fit(X_train_tfidf, y_train)
#y_pred = clf.predict(X_test_tfidf)
#metrics.accuracy_score(y_test, y_pred) # 0.6165863819815128
# GRIDSEARCH
from sklearn.model_selection import GridSearchCV
# prepare a range of values to test
parameters = {'alpha': [0.01,0.1,1,1.5,2]}
# create and fit a ridge regression model, testing each alpha
model = MultinomialNB()
grid = GridSearchCV(estimator=model, param_grid=parameters)
grid.fit(X_train_dtm, y_train)
# summarize the results of the grid search
print(grid.best_score_) # 0.661299422768109
print(grid.best_estimator_.alpha) # 1
grid.fit(X_dtm, y) # 0.6498925851181564 - lower??