DeepBioSim/sim_sample_old.py at master · biocoms/DeepBioSim · GitHub

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
# comp_ds.py
import os
import glob
import pandas as pd
import torch
from torch.utils.data import TensorDataset, DataLoader
import torch.optim as optim
import numpy as np
from scipy.stats import ttest_ind
from KDEpy import FFTKDE
from models.vae import VAE, train as train_vae, k_fold_validation_vae
from models.iwae import IWAE, train as train_iwae, k_fold_validation_iwae
from models.kde_mcmc import mcmc_sampling
from mks_test import mkstest
from utils import plot_pca, plot_tsne, svd_reduce, svd_reconstruct, shannon

import pdb

# --------------------
# Configuration
# --------------------
latent_dim = 8
hidden_dim = 64
num_epochs = 800
batch_size = 128
learning_rate = 1e-3
K = 20
random_seed = 42

torch.manual_seed(random_seed)
np.random.seed(random_seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


def process_file(filepath: str):
    dataset_name = os.path.splitext(os.path.basename(filepath))[0]
    print(f"\n=== Processing {dataset_name} ===")

    # Load data (first column as index, first row as header)
    df = pd.read_csv(filepath, index_col=0)
    # data = df.values.T  # shape: (n_samples, n_features)
    data = np.log1p(df.values)
    n_samples, input_dim = data.shape
    Z, V_red = svd_reduce(data, max_fraction=0.1)
    input_dim_red = Z.shape[1]
    print(f"Data: {n_samples} samples, {input_dim} features")

    loader = DataLoader(
        TensorDataset(torch.from_numpy(Z).float()), batch_size=batch_size, shuffle=True
    )
    print(f"Loaded reduced data: {n_samples} samples, {input_dim_red} features")

    # ----- Vanilla VAE -----
    vae = VAE(input_dim_red, hidden_dim, latent_dim).to(device)
    opt_vae = optim.Adam(vae.parameters(), lr=learning_rate)
    train_vae(vae, loader, opt_vae, num_epochs=num_epochs, device=device)
    vae.eval()
    with torch.no_grad():
        z = torch.randn(n_samples, latent_dim, device=device)
        gen_vae_red = vae.decode(z).cpu().numpy()
    gen_vae = svd_reconstruct(gen_vae_red, V_red)
    plot_pca(data.T, gen_vae.T, "VAE", dataset_name)
    # plot_tsne(data, gen_vae, "VAE", dataset_name)

    # ----- IWAE -----
    iwae = IWAE(input_dim_red, latent_dim, hidden_dim, K).to(device)
    opt_iwae = optim.Adam(iwae.parameters(), lr=learning_rate)
    # train and print avg loss per epoch
    train_iwae(iwae, loader, opt_iwae, num_epochs=num_epochs, device=device)
    iwae.eval()
    with torch.no_grad():
        z = torch.randn(n_samples, latent_dim, device=device)
        gen_iwae_red = iwae.sample(n_samples)
    gen_iwae = svd_reconstruct(gen_iwae_red, V_red)
    plot_pca(data.T, gen_iwae.T, "IWAE", dataset_name)
    # plot_tsne(data, gen_iwae, "IWAE", dataset_name)

    # ----- KDE-MCMC -----
    # kde = FFTKDE(kernel="gaussian").fit(Z)
    # bw = kde.bw
    # init = Z[np.random.choice(n_samples)]
    # gen_mcmc_red = mcmc_sampling(
    #     initial_point=init,
    #     data=Z,
    #     bandwidth=bw,
    #     num_samples=n_samples,
    #     step_size=0.5,
    #     leapfrog_steps=10,
    #     burn_in=100,
    # )
    # gen_mcmc = svd_reconstruct(gen_mcmc_red, V_red)
    # plot_pca(data, gen_mcmc, "KDE-MCMC", dataset_name)
    # plot_tsne(data, gen_mcmc, "KDE-MCMC", dataset_name)

    # NOTE mkstest takes a long time to run, and the library itself is a bit buggy

    # print("VAE mkstest:")
    # mkstest(gen_vae, data, alpha=0.05, verbose=True)
    # print("IWAE mkstest:")
    # mkstest(gen_iwae, data, alpha=0.05, verbose=True)
    # print("KDE-MCMC mkstest:")
    # mkstest(gen_mcmc, data, alpha=0.05, verbose=True)

    # NOTE shannon diversity can't compute negative values so it's not included in the analysis
    # H_orig = shannon(data)
    # H_vae = shannon(gen_vae)
    # H_iwae = shannon(gen_iwae)
    # H_mcmc = shannon(gen_mcmc)

    # t_vae, p_vae = ttest_ind(H_orig, H_vae)
    # t_iwae, p_iwae = ttest_ind(H_orig, H_iwae)
    # t_mcmc, p_mcmc = ttest_ind(H_orig, H_mcmc)

    # print("Shannon diversity t-test results:")
    # print(f" VAE:  t={t_vae:.3f}, p={p_vae:.3e}")
    # print(f" IWAE: t={t_iwae:.3f}, p={p_iwae:.3e}")
    # print(f" MCMC: t={t_mcmc:.3f}, p={p_mcmc:.3e}")

    # means (axis=1 mean for each gene)
    # mean_orig = np.mean(data, axis=0)
    # mean_vae = np.mean(gen_vae, axis=0)
    # mean_iwae = np.mean(gen_iwae, axis=0)
    # mean_mcmc = np.mean(gen_mcmc, axis=0)

    # t-tests on means
    # t_mean_vae, p_mean_vae = ttest_ind(mean_orig, mean_vae)
    # t_mean_iwae, p_mean_iwae = ttest_ind(mean_orig, mean_iwae)
    # t_mean_mcmc, p_mean_mcmc = ttest_ind(mean_orig, mean_mcmc)

    # print("Mean t-test results:")
    # print(f" VAE:  t={t_mean_vae:.3f}, p={p_mean_vae:.3e}")
    # print(f" IWAE: t={t_mean_iwae:.3f}, p={p_mean_iwae:.3e}")
    # print(f" MCMC: t={t_mean_mcmc:.3f}, p={p_mean_mcmc:.3e}")

    # medians
    # med_orig = np.median(data, axis=0)
    # med_vae = np.median(gen_vae, axis=0)
    # med_iwae = np.median(gen_iwae, axis=0)
    # med_mcmc = np.median(gen_mcmc, axis=0)

    # from scipy.stats import wilcoxon

    # paired Wilcoxon signed-rank tests on medians
    # w_med_vae, p_w_med_vae = wilcoxon(med_orig, med_vae)
    # w_med_iwae, p_w_med_iwae = wilcoxon(med_orig, med_iwae)
    # if you had paired KDE-MCMC medians too:
    # w_med_mcmc, p_w_med_mcmc = wilcoxon(med_orig, med_mcmc)

    # print("Wilcoxon signed-rank test on medians:")
    # print(f" VAE:  W={w_med_vae:.3f}, p={p_w_med_vae:.3e}")
    # print(f" IWAE: W={w_med_iwae:.3f}, p={p_w_med_iwae:.3e}")
    # print(f" MCMC: W={w_med_mcmc:.3f}, p={p_w_med_mcmc:.3e}")

    # from scipy.stats import mannwhitneyu

    # u_med_vae, p_u_med_vae = mannwhitneyu(med_orig, med_vae)
    # u_med_iwae, p_u_med_iwae = mannwhitneyu(med_orig, med_iwae)

    # print("Mann–Whitney U test on medians:")
    # print(f" VAE:  W={u_med_vae:.3f}, p={p_u_med_vae:.3e}")
    # print(f" IWAE: W={u_med_iwae:.3f}, p={p_u_med_iwae:.3e}")
    # print(f" MCMC: W={w_med_mcmc:.3f}, p={p_w_med_mcmc:.3e}")


# NOTE TCGA will *not* run on MCMC because of the high dimensionality
if __name__ == "__main__":
    os.makedirs("./output", exist_ok=True)
    # process_file("./input/momspi16s.csv")
    # process_file("./input/t2d16s.csv")
    process_file("./input/ibd.csv")
    process_file("./input/vaginal.csv")
    # process_file("./input/TCGA_HNSC_rawcount_data_t.csv")