-
Notifications
You must be signed in to change notification settings - Fork 1
/
Regression.py
183 lines (165 loc) · 7.74 KB
/
Regression.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
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
from utils import *
from sklearn.utils import Bunch
from scipy.stats import spearmanr
class LinReg:
"""
Gather data from file, extract additional features and labels, later use to train a linear regressor
"""
def __init__(self, **kwargs):
filename = kwargs.get('filename', None)
drop_wins = kwargs.get('drop_wins', False)
data = kwargs.get('data', None)
label = kwargs.get('label', None)
self.X = data
self.Y = label
if data is None:
tmp_data = pd.read_excel(filename, index_col='Gene index')
self.str_seriesses = tmp_data[['PA', 'UTR5', 'ORF']]
self.X = tmp_data.drop(columns=['PA', 'UTR5', 'ORF', 'argenin frequnecy '])
self.add_features()
self.Y = tmp_data['PA']
self.normalized_Y = self.Y.apply(lambda x: ((x - min(self.Y)) / (max(self.Y) - min(self.Y))) ** 0.1)
if drop_wins: self.X = self.X.drop(columns=[name for name in self.X.columns if 'dow' in name])
[self.model, self.train_test_data] = self.get_model(**kwargs)
def add_features(self):
orf_cols = ['orf_Arg_freq', 'orf_Ala_freq', 'orf_Gly_freq', 'orf_Val_freq']
orf_aa_freqs = add_aas_freq(self.str_seriesses['ORF'], orf_cols, ['R', 'A', 'G', 'V'])
self.X = pd.concat([self.X, orf_aa_freqs], axis=1)
self.X['free_var'] = np.ones(len(self.X))
self.X['avg_win'] = np.mean(self.X.iloc[:, 4:104], axis=1)
self.X['std_win'] = np.std(self.X.iloc[:, 4:104], axis=1)
self.X['ORF_len'] = self.str_seriesses['ORF'].apply(len)
self.X['UTR_len'] = self.str_seriesses['UTR5'].apply(len)
self.X['TATA_loc'] = self.str_seriesses['UTR5'].apply(get_tata_loc)
self.X['GC_count'] = self.str_seriesses['UTR5'].apply(gc_count)
self.X['num_start_codons'] = self.str_seriesses['ORF'].apply(num_start_codons)
for nuc in ['A', 'T', 'G', 'C']: self.X["utr_{0}_freq".format(nuc)] = get_nuc_freq(nuc,
self.str_seriesses['UTR5'])
for nuc in ['A', 'T', 'G', 'C']: self.X["orf_{0}_freq".format(nuc)] = get_nuc_freq(nuc,
self.str_seriesses['ORF'])
def plot_col_num(self, col_num):
"""
this method scatter plots the feature and label plot
:param col_num: data frame column number
"""
col = self.X.iloc[:, col_num]
plt.figure()
plt.scatter(self.X[col], self.Y)
plt.xlabel(col), plt.ylabel('PA')
def get_model(self, **kwargs):
"""
Train a linear regressor based on inside or outside data and return model and data split
"""
# TODO: cross-validation (k-fold)
train_test_data = kwargs.get('train_test_data', None)
test_size = kwargs.get('test_size', 0.3)
normalize = kwargs.get('normalize', False)
if train_test_data is None:
train_test_data = Bunch()
y = self.normalized_Y if normalize else self.Y
train_test_data.x_train, train_test_data.x_test, \
train_test_data.y_train, train_test_data.y_test = \
train_test_split(self.X, y, test_size=test_size)
reg = LinearRegression()
reg.fit(train_test_data.x_train, train_test_data.y_train)
return reg, train_test_data
def asses_model(self, data=None, Training=False,prtf = True):
"""
Asses model performance and print results
Training = use training data instead of test data
"""
reg = self.model
x_test = self.train_test_data.x_test
y_test = self.train_test_data.y_test
if Training:
x_test = self.train_test_data.x_train
y_test = self.train_test_data.y_train
if data:
x_test = data.x
y_test = data.y
y_pred = pd.Series(reg.predict(x_test), index=y_test.index)
if prtf:
print("R^2 for data is:{0}".format(reg.score(x_test, y_test)))
print(spearmanr(y_pred, y_test))
return spearmanr(y_pred, y_test)
def visualize_model_performance(self, data=None):
"""Plot model prediction vs label"""
x_test = self.train_test_data.x_test
y_test = self.train_test_data.y_test
if data: # Allows performance comparison between models on same dataset
x_test = data.x
y_test = data.y
y_pred = pd.Series(self.model.predict(x_test), index=y_test.index)
plt.figure()
y_pred_reind = y_pred.copy()
y_test_reind = y_test.copy()
y_pred_reind.index = range(len(y_pred))
y_test_reind.index = range(len(y_test))
asc_ind = y_test_reind.sort_values().index
plt.scatter(range(len(y_pred)), y_pred_reind.iloc[asc_ind], label="y_pred")
plt.scatter(range(len(y_test)), y_test_reind.iloc[asc_ind], label="y_test")
plt.xticks(asc_ind)
plt.legend()
plt.title('Test PA')
plt.show()
def get_coefs(self, n=10, pltf=False, prtf = True):
"""
return n primary model coefficients
:pltf is plot_flag
"""
mdl = self.model
abscoefs = abs(mdl.coef_)
if len(mdl.coef_) < n:
n = len(mdl.coef_) - 1
if pltf:
plt.figure(), plt.bar(np.arange(len(abscoefs)).ravel(), abscoefs)
plt.title('Regression coefficients values')
plt.xticks(range(len(self.X.columns)), self.X.columns)
sorted_idx = np.flip(abscoefs.argsort())
sorted_features = self.X.columns[sorted_idx]
if prtf:
for i in range(n):
print("feature name:{0}, feature coeff:{1}".format(sorted_features[i], abscoefs[sorted_idx][i]))
return mdl.coef_
class RidgeReg(LinReg):
def __init__(self, **kwargs):
LinReg.__init__(self, **kwargs)
[self.model, self.train_test_data] = self.get_model(**kwargs)
def get_model(self, **kwargs):
"""
Train a linear regressor based on inside or outside data and return model and data split
"""
train_test_data = kwargs.get('train_test_data', None)
test_size = kwargs.get('test_size', 0.3)
normalize = kwargs.get('normalize', False)
if train_test_data is None:
train_test_data = Bunch()
y = self.normalized_Y if normalize else self.Y
train_test_data.x_train, train_test_data.x_test, \
train_test_data.y_train, train_test_data.y_test = \
train_test_split(self.X, y, test_size=test_size)
reg = Ridge()
reg.fit(train_test_data.x_train, train_test_data.y_train)
return reg, train_test_data
if __name__ == "__main__":
lr = LinReg(filename="Known_set_Bacillus.xlsx", drop_wins=True)
best_features = sfs(round(len(lr.X) ** 0.5), lr.X, lr.Y)
# This part is ugly but not sure how to solve this 'cause pandas are fucking mutable TODO: find better solution
best_data = Bunch()
best_data.x_train = lr.train_test_data.x_train[list(best_features)].copy()
best_data.x_test = lr.train_test_data.x_test[list(best_features)].copy()
best_data.y_train = lr.train_test_data.y_train.copy()
best_data.y_test = lr.train_test_data.y_test.copy()
# End of ugly part
only_best = LinReg(data=lr.X[list(best_features)], label=lr.Y, train_test_data=best_data)
only_best.visualize_model_performance()
print('***All features results***')
print('--training results--')
lr.asses_model(Training=True)
print('--test results--')
lr.asses_model()
print('***Only best features results***')
print('--training results--')
only_best.asses_model(Training=True)
print('--test results--')
only_best.asses_model()