contribution_plot.py

import matplotlib

matplotlib.use('Agg')
import matplotlib.pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
import tensorflow as tf
import numpy as np
from tensorflow.python.keras.models import Sequential, Model
from deepexplain.tensorflow import DeepExplain
import os
from load_data import load_data

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"  # see issue #152
os.environ["CUDA_VISIBLE_DEVICES"] = "5"


def retrack(model_loc):
    model = tf.keras.models.load_model(model_loc)
    model.summary()
    print('///////////////////////////')

    with DeepExplain(session=tf.compat.v1.keras.backend.get_session()) as de:
        input_tensor = model.input
        fModel = Model(inputs=input_tensor, outputs=model.output)
        target_tensor = fModel(input_tensor)

        attributions_pos = de.explain('grad*input', target_tensor, input_tensor, xs=x_val[:100], ys=y_val[:100])

    return attributions_pos


def plotpdf(location, weight):
    print('---------------------')
    print(sum(weight)[:, 1])
    print(sum(weight)[:, 2])
    print(sum(weight)[:, 3])
    print(sum(weight)[:, 0])
    a = sum(weight)[:, 0]
    b = sum(weight)[:, 1]
    c = sum(weight)[:, 2]
    d = sum(weight)[:, 3]
    # print(type(a))
    print(len(a))
    all = np.concatenate((a, b, c, d)).tolist()
    # print(type(all))
    # print(all)
    import heapq
    re1 = map(all.index, heapq.nlargest(20, all))
    re2 = heapq.nlargest(20, all)
    re3 = map(all.index, heapq.nsmallest(20, all))
    print(list(re1))
    print(list(re3))
    # print(re2)

    weight[:, 249:251, :] = 0
    print('---------')

    with PdfPages(location) as pdf:
        label = ['A', 'C', 'G', 'T']
        for i in range(4):
            plt.title(label[i])
            plt.bar(np.arange(501) - 250, sum(weight)[:, i])
            pdf.savefig()
            plt.close()
            plt.title(label[i])
            plt.bar(np.arange(101) - 50, sum(weight)[225:326, i])
            pdf.savefig()
            plt.close()

    print('DONE!!!!!!!')


if __name__ == '__main__':
    data_path = '/home/yuxuan/dp/longer_seq_data/eif3a_Full_250.csv'
    x_train, x_test, x_val, y_test, y_train, y_val = load_data(data_path)
    attributions_pos = retrack('/home/yuxuan/dp/model/eif3a_Full_250_CRNNmodel.h5')
    print('---------BEFORE THE PLOTDF')
    plotpdf('/home/yuxuan/dp_m6a_org/eif3a_full_501.pdf', attributions_pos)