-
Notifications
You must be signed in to change notification settings - Fork 1
/
active_learning_drug_discovery.py
142 lines (126 loc) · 6.03 KB
/
active_learning_drug_discovery.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
import sys
import random
import numpy as np
import pickle
from sklearn.ensemble import RandomForestClassifier
from sklearn.cross_validation import train_test_split
import pandas as pd
import matplotlib.pyplot as plt
def get_features_and_responses(data):
#return x, y = features, response
x = data[data.columns[1:]]
y = []
for i in data['?']:
if i == "I":
y.append((1,0))
else:
y.append((0,1))
return x, y
def get_model_performance(data, batch_size, strategy, seed):
#return a tuple [%chemical_space, %hit_discovery] for each batch iterations
print "Loading Features:"
x, y = get_features_and_responses(data)
print "Spliting the data:"
x_train, x_test, y_train, y_test, indices_train, indices_test = train_test_split(x,y,range(data.shape[0]), test_size=1-(float(batch_size)/100), random_state = seed)
print "Starting iterations:"
total_test_space = data.shape[0]*(1-(float(batch_size)/100))
picks = (batch_size * data.shape[0])/100 #number of picks of small molecules at each iteration
hits = len([item for item in y_train if item[1] == 1]) #hits stores number of hits for each iteraction
total_hits = len([item for item in y if item[1] == 1]) #total small molecules hits for a given target
result = []
#measure performance
percent_chemical_space = int(100-(float(len(indices_test))/total_test_space)*100)
percent_hits = (hits*100)/total_hits
result.append([percent_chemical_space,percent_hits])
itr = 1
print "Iterations: ", itr, strategy.__name__, "|Percent chemical space: ", percent_chemical_space, "|Percent hits: ", percent_hits, "|hits: ", hits
while len(indices_test) > 0:
indices_train, indices_test = active_learning(x_train, y_train, x_test, y_test, indices_train, indices_test, strategy, picks,seed)
#new x_train, y_train, and x_test
x_train = x.ix[indices_train]
y_train = [y[item] for item in indices_train]
x_test = x.ix[indices_test]
y_test = [y[item] for item in indices_test]
picks = min(len(y_test), picks)
hits = len([item for item in y_train if item[1] == 1])
itr += 1
#measure performance
percent_chemical_space = int(100-(float(len(indices_test))/total_test_space)*100)
percent_hits = (hits*100)/total_hits
result.append([percent_chemical_space,percent_hits])
print "Iterations: ", itr, strategy.__name__, "|Percent chemical space: ", percent_chemical_space, "|Percent hits: ", percent_hits, "|hits: ", hits
print "------"
return result
def active_learning(x_train,y_train,x_test, y_test, indices_train, indices_test, strategy, picks,seed):
#return the indices of new train and test data according to strategy
if strategy.__name__ == "get_optimal_sorted_indices":
indices_picked = get_optimal_sorted_indices(y_test, indices_test)[:picks]
else:
df = RandomForestClassifier(n_estimators=12, criterion='gini', bootstrap=True, oob_score=False, n_jobs=4, random_state = seed, verbose=0, class_weight='auto')
df.fit(x_train,y_train)
if strategy.__name__ == "get_max_entropy_sorted_indices":
tup = df.predict_proba(x_test)[1]
elif strategy.__name__ == "get_max_prob_sorted_indices":
tup = df.predict_proba(x_test)[0]
else:
tup = df.predict(x_test)
indices_picked = strategy(tup, indices_test)[:picks]
hits = len([item for item in y_train if item[1] == 1])
indices_test = [item for item in indices_test if item not in indices_picked]
indices_train.extend(indices_picked)
return indices_train, indices_test
def get_random_indices(tuples, indices):
#return indices of test_set with random shuffle
random.seed(0)
random.shuffle(indices)
return indices
def get_max_entropy_sorted_indices(tuples, indices):
#return indices of test_set according to max entropy - decending
entropy = []
for i in tuples:
entropy.append((i[0] * np.log(i[0])) + (i[1] * np.log(i[1])))
sorted_index = np.argsort(entropy)
return np.array(indices)[sorted_index]
def get_max_prob_sorted_indices(tuples, indices):
#return indices of test_set according to max prob - decending
prob = []
for i in tuples:
prob.append(i[1])
sorted_index = np.argsort(prob)
return np.array(indices)[sorted_index]
def get_optimal_sorted_indices(tuples, indices):
#return indices of test_set according to optimal picks
optimal = []
for i in tuples:
optimal.append(i[1])
sorted_index = np.argsort(optimal)[::-1]
return np.array(indices)[sorted_index]
def plot_model_comparision(random_tup1, max_entropy_tup2, max_prob_tup3, optimal_tup4, outfile):
#output a png file displaying the performance of different strategies
random_res1 = zip(*random_tup1)
max_entropy_res2 = zip(*max_entropy_tup2)
max_prob_res3 = zip(*max_prob_tup3)
optimal_res4 = zip(*optimal_tup4)
#plot lines
plt.plot(random_res1[0],random_res1[1], "k--", label = "Random")
plt.plot(max_entropy_res2[0],max_entropy_res2[1],'bo-', label = "Max Entropy")
plt.plot(max_prob_res3[0], max_prob_res3[1],'go-', label = "Max Probability")
plt.plot(optimal_res4[0],optimal_res4[1],'r--', label = "Optimal")
#add legend
legend = plt.legend(loc="lower right")
frame = legend.get_frame()
#label axis
plt.xlabel("Tested Chemical Space (%)")
plt.ylabel("Discovered Hits (%)")
#save figure
plt.savefig(outfile)
print "Loading pickle data object"
#data = pd.read_csv(sys.argv[1])
#data.to_pickle("drug_discovery_data")
data = pd.read_pickle(sys.argv[1])
random_tup1 = get_model_performance(data, 5, get_random_indices, 0)
max_entropy_tup2 = get_model_performance(data, 5, get_max_entropy_sorted_indices, 1)
max_prob_tup3 = get_model_performance(data, 5, get_max_prob_sorted_indices, 1)
optimal_tup4 = get_model_performance(data, 5, get_optimal_sorted_indices, 0)
outfile = "performance_active_learner_comparision.png"
plot_model_comparision(random_tup1, max_entropy_tup2, max_prob_tup3, optimal_tup4, outfile)