-
Notifications
You must be signed in to change notification settings - Fork 0
/
Test03_BeatAML.py
223 lines (168 loc) · 8.95 KB
/
Test03_BeatAML.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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 14 17:28:20 2020
@author: SRDhruba
"""
#!/usr/bin/env python
# coding: utf-8
# In[ ]:
import os
import numpy as np
import pandas as pd
import pickle
from time import time
from tqdm import tqdm
# from skrebate import ReliefF
from sklearn.preprocessing import StandardScaler
# from sklearn.model_selection import KFold
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor, AdaBoostRegressor
from sklearn.svm import SVR as SupportVectorRegressor
from sklearn.linear_model import Lasso, LassoCV, ElasticNet
from sklearn.neighbors import KNeighborsRegressor
from scipy.stats import pearsonr, spearmanr
## Path & FILE...
PATH = "%s\\Google Drive\\Study\\ECE 5332-009 - Topics in EE, Data Science\\BeatAML\\" % os.getenv("HOMEPATH")
DIR = ("Training\\", "Leaderboard\\")
FILE = ("rnaseq.csv", "dnaseq.csv", "clinical_numerical.csv", "clinical_categorical.csv",
"clinical_categorical_legend.csv", "aucs.csv", "response.csv")
os.chdir(PATH)
### Training data...
RNA_TR = pd.read_csv(PATH + DIR[0] + FILE[0], header = 0)
# CLN_TR = pd.read_csv(PATH + DIR[0] + FILE[2], header = 0)
# CLC_TR = pd.read_csv(PATH + DIR[0] + FILE[3], header = 0)
AUC_TR = pd.read_csv(PATH + DIR[0] + FILE[5], header = 0)
## Leaderboard data...
RNA_LB = pd.read_csv(PATH + DIR[1] + FILE[0], header = 0)
# CLN_LB = pd.read_csv(PATH + DIR[1] + FILE[2], header = 0)
# CLC_LB = pd.read_csv(PATH + DIR[1] + FILE[3], header = 0)
AUC_LB = pd.read_csv(PATH + DIR[1] + FILE[5], header = 0)
all(AUC_TR.inhibitor.unique() == AUC_LB.inhibitor.unique()) ## Check if same drugs
all(RNA_TR[["Gene", "Symbol"]] == RNA_LB[["Gene", "Symbol"]]) ## Check if same genes
lab_id_list = dict(TR = RNA_TR.columns, LB = RNA_LB.columns)
gene_list, drug_list = RNA_TR[["Gene", "Symbol"]], AUC_TR.inhibitor.unique().tolist()
## Preprocessing...
RNA_TR.index, RNA_LB.index = gene_list.Symbol, gene_list.Symbol
RNA_TR, RNA_LB = RNA_TR.iloc[:, 2:], RNA_LB.iloc[:, 2:]
# print(RNA_TR.shape, RNA_LB.shape)
var_idx = (-RNA_TR.var(axis = 1)).to_numpy().argsort()[:40000] ## Filter by gene variability
RNA_TR_filt, RNA_LB_filt = RNA_TR.iloc[var_idx, :], RNA_LB.iloc[var_idx, :]
# In[ ]:
## Scaling...
zscore = lambda data, ref_data: StandardScaler().fit(ref_data).transform(data)
## Feature selection...
def LassoFS(X, y, seed = 0):
# FS = LassoCV(fit_intercept = True, normalize = False, n_alphas = 100, tol = 1e-3, cv = 5, selection = "random",
# random_state = seed, n_jobs = -1)
# FS = Lasso(fit_intercept = True, normalize = False, alpha = 0.001, tol = 1e-3, selection = "random", random_state = seed)
FS = Lasso(fit_intercept = True, normalize = False, alpha = 0.001, selection = 'random', random_state = seed, max_iter = 3000)
features = (FS.fit(X, y).coef_ != 0).nonzero()[0]
return features
## Predictive models...
def RF(X_train, y_train, X_test, seed = 0):
mdl = RandomForestRegressor(n_estimators = 200, criterion = "mae", min_samples_leaf = 5, random_state = seed, n_jobs = -1)
y_pred = mdl.fit(X_train, y_train).predict(X_test)
return y_pred
def SVR(X_train, y_train, X_test):
mdl = SupportVectorRegressor(kernel = "poly", degree = 3, coef0 = 1.0, gamma = "scale", tol = 1e-3, C = 10, max_iter = 2000)
y_pred = mdl.fit(X_train, y_train).predict(X_test)
return y_pred
def EN(X_train, y_train, X_test, seed):
mdl = ElasticNet(fit_intercept = True, l1_ratio = 0.005, alpha = 0.2, tol = 1e-3, max_iter = 2000, selection = "random",
random_state = seed)
y_pred = mdl.fit(X_train, y_train).predict(X_test)
return y_pred
def KNN(X_train, y_train, X_test):
mdl = KNeighborsRegressor(n_neighbors = 7, weights = "distance", algorithm = "auto", metric = "minkowski", p = 1, n_jobs = -1)
y_pred = mdl.fit(X_train, y_train).predict(X_test)
return y_pred
def ADB(X_train, y_train, X_test, seed = 0, base = "tree"):
if base == "tree":
base = DecisionTreeRegressor(criterion = "mae", max_depth = 3, min_samples_leaf = 5, max_features = "sqrt",
min_impurity_decrease = 1e-6, splitter = "best", random_state = seed)
elif base == "linear":
base = ElasticNet(fit_intercept = True, l1_ratio = 0.005, alpha = 0.2, tol = 1e-3, max_iter = 2000, selection = "random",
random_state = seed)
#### Choice loop ends...
mdl = AdaBoostRegressor(base_estimator = base, n_estimators = 150, learning_rate = 0.8, loss = "exponential", random_state = seed)
y_pred = mdl.fit(X_train, y_train).predict(X_test)
return y_pred
## Evaluate model performance...
def EVAL_PERF(y_label, y_pred, alpha = 0.05):
y_label, y_pred = np.array(y_label).squeeze(), np.array(y_pred).squeeze()
PCC, pval = pearsonr(y_label, y_pred); #PCC = PCC if pval < alpha else 0
SCC, pval = spearmanr(y_label, y_pred); #SCC = SCC if pval < alpha else 0
NRMSE = np.sqrt(((y_label - y_pred)**2).mean()) / y_label.std(ddof = 0)
NMAE = (np.abs(y_label - y_pred)).mean() / (np.abs(y_label - y_label.mean())).mean()
return PCC, SCC, NRMSE, NMAE
# In[ ]
### TEST ON LEADERBOARD DATA...
FS_switch = "ReliefF"
if FS_switch == "ReliefF":
p_top = 15000
# feature_ranks = {kk: [ ] for kk in drug_list}
with open("FS_Drugs_122_ReliefF_3_folds.pickle", "rb") as file:
feature_ranks = pickle.load(file)
elif FS_switch == "Lasso":
feature_ranks = dict.fromkeys(drug_list)
####
AUC_LB_pred = { }; metrics = ["PCC", "SCC", "NRMSE", "NMAE"]
models = ["RF", "SVR", "EN", "KNN", "ADB", "ENS", "ENS1234", "ENS1235", "ENS1245", "ENS1345", "ENS2345", "ENS123", "ENS124", "ENS125",
"ENS134", "ENS135", "ENS145", "ENS234", "ENS235", "ENS245", "ENS345"]
RESULTS_LB = {MM: pd.DataFrame(dtype = float, index = drug_list, columns = metrics) for MM in models}
N = len(drug_list); count = 0
dt = time()
for drug in tqdm(drug_list[:N]):
count += 1; # print("\nChosen drug# = %d: %s" % (count, drug))
y_TR = AUC_TR.iloc[(AUC_TR.inhibitor == drug).tolist(), :]
y_LB = AUC_LB.iloc[(AUC_LB.inhibitor == drug).tolist(), :]
X_TR, y_TR = RNA_TR_filt.loc[:, y_TR.lab_id].T, y_TR["auc"].to_numpy()
X_LB, y_LB = RNA_LB_filt.loc[:, y_LB.lab_id].T, y_LB["auc"].to_numpy()
Y_LB_pred = pd.DataFrame(dtype = float, index = X_LB.index, columns = ["Actual"] + models)
fold = 0; # dt = time()
# ## Perform ReliefF...
# dt = time(); FS.fit(X_TR.values, y_TR); dt = time() - dt; print("Elapsed time = %0.4f sec." % dt)
# feat_top = FS.top_features_; feature_ranks[drug].append(feat_top)
## From saved file...
if FS_switch == "ReliefF":
feat_top_set = feature_ranks[drug][fold]
feat_top = feat_top_set[:p_top]
elif FS_switch == "Lasso":
feat_top = LassoFS(X_TR, y_TR, seed = 2020)
feature_ranks[drug] = feat_top
X_TR, X_LB = X_TR.iloc[:, feat_top], X_LB.iloc[:, feat_top]
## Perform prediction...
Y_LB_pred.loc[:, "Actual"] = y_LB
Y_LB_pred.loc[:, "RF"] = RF(X_TR, y_TR, X_LB, seed = 0)
Y_LB_pred.loc[:, "SVR"] = SVR(X_TR, y_TR, X_LB)
Y_LB_pred.loc[:, "EN"] = EN(X_TR, y_TR, X_LB, seed = 0)
Y_LB_pred.loc[:, "KNN"] = KNN(X_TR, y_TR, X_LB)
Y_LB_pred.loc[:, "ADB"] = ADB(X_TR, y_TR, X_LB, seed = 0)
RESULTS_LB["RF"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["RF"])
RESULTS_LB["SVR"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["SVR"])
RESULTS_LB["EN"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["EN"])
RESULTS_LB["KNN"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["KNN"])
RESULTS_LB["ADB"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["ADB"])
# fold += 1
# dt = time() - dt; print("Elapsed time = %0.4f sec." % dt)
## Ensemble prediction...
Y_LB_pred.loc[:, "ENS"] = Y_LB_pred.loc[:, np.array(models)[:5]].mean(axis = 1)
RESULTS_LB["ENS"].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred["ENS"])
for MM in models[6:]:
Y_LB_pred.loc[:, MM] = Y_LB_pred.loc[:, map(lambda i: models[int(i)-1], MM.split("ENS")[1])].mean(axis = 1)
RESULTS_LB[MM].loc[drug, :] = EVAL_PERF(Y_LB_pred["Actual"], Y_LB_pred[MM])
#### Ensemble loop ends...
AUC_LB_pred[drug] = Y_LB_pred
#### Whole loop ends...
dt = time() - dt; print("Elapsed time = %0.4f sec." % dt)
RESULTS_LB_MEAN = pd.DataFrame({MM: RESULTS_LB[MM].mean(axis = 0) for MM in models})
RESULTS_LB["Mean"] = RESULTS_LB_MEAN.T
print("Mean performance for %d inhibitors = \n" % N, RESULTS_LB_MEAN.T)
#%%
# from sklearn.linear_model import LassoCV
# FS = Lasso(fit_intercept = False, normalize = False, alpha = 0.001, tol = 1e-3,
# selection = "random", random_state = 0)
# # FS = LassoCV(fit_intercept = True, normalize = True, tol = 1e-3, n_alphas = 200, cv = 5,
# # selection = "random", random_state = 0)
# aa = FS.fit(X_TR, y_TR).coef_
# print(sum(aa != 0))