pythonのプログラムのメモリについて

pythonのプログラムを実行していく上でメモリについてわからないことがありましたので質問させていただきます。
現在、pythonでプログラムを書いているのですが、おそらく自分のプログラムの構成の仕方が悪いせいで、メモリを多く使ってしまっています。
そこで、なるべくメモリを消費しないようなpythonのプログラムを作成したいと思っています。

以下が現在使っているpythonのプログラムコードとなっています。(長くなってしまい、読みづらかったらすみません。)

python
1# -*- coding: utf-8 -*-
2#@test {"output": "ignore"}
3import tensorflow as tf
4import numpy as np
5import pandas as pd
6import glob
7import os
8import math
9
10def read_file(filename):
11    ds_tmp = []
12    with open(filename, 'r') as f:
13        print(filename)
14        for i in f:
15            txt = i.split()
16            txt = [ float(s) for s in txt]
17            ds_tmp.append(txt)
18            
19        ds = np.array(ds_tmp)
20        return ds
21    
22
23def process(label):
24    train_filename = "../text/text2/input_train_{}.txt".format(label)
25    test_filename = "../text/text2/input_test_{}.txt".format(label)
26    train_data = read_file(train_filename)
27    test_data = read_file(test_filename)
28    y_train = train_data[:, 4:]/50.0 # 4列目のみを抽出
29    x_train = train_data[:, :4] # 0~4列目を抽出
30    y_test = test_data[:, 4:]/50.0
31    x_test = test_data[:, :4]
32
33    num_examples = y_train.shape[0]
34    num_input = x_train.shape[1]
35    
36    bias_with_x = np.insert(x_train, 0, 1, axis=1).astype(np.float32)
37    bias_with_x_test = np.insert(x_test, 0, 1, axis=1).astype(np.float32)
38    
39    num_hidden1 = 40
40    num_hidden2 = 40
41    training_steps = 25000
42    learning_rate = 0.005
43    lambda_2 = 0.00000001
44    MAE_train_ave = 0
45    MAE_test_ave = 0
46
47    for seed in range(1,11):
48        losses = []
49
50        with tf.Session() as sess:
51            # Set up all the tensors, variables, and operations.
52            input = tf.constant(bias_with_x)
53            target = tf.constant(np.transpose([y_train]).astype(np.float32))
54            shape = [num_input+1, num_hidden1]
55            n = shape[0] * shape[1]
56            stddev = math.sqrt(2.0 / n)
57            initial = tf.truncated_normal(shape=shape, stddev=stddev, seed=seed)
58            weights = tf.Variable(initial, name=None)
59            bias = tf.Variable(tf.random_uniform([num_hidden1], minval=-2.0, maxval=2.0, seed=seed))
60
61            shape = [num_hidden1, num_hidden2]
62            n = shape[0] * shape[1]
63            stddev = math.sqrt(2.0 / n)
64            initial = tf.truncated_normal(shape=shape, stddev=stddev, seed=seed)
65            weights1 = tf.Variable(initial, name=None)
66            bias1 = tf.Variable(tf.random_uniform([num_hidden2], minval=-2.0, maxval=2.0, seed=seed))
67
68            shape = [num_hidden2, 1]
69            n = shape[0] * shape[1]
70            stddev = math.sqrt(2.0 / n)
71            initial = tf.truncated_normal(shape=shape, stddev=stddev, seed=seed)
72            weights2 = tf.Variable(initial, name=None)
73            bias2 = tf.Variable(tf.random_uniform([1], minval=-2.0, maxval=2.0, seed=seed))
74            tf.global_variables_initializer().run()
75
76            y1 = tf.nn.relu(tf.matmul(input, weights))
77            y2 = tf.nn.relu(tf.matmul(y1, weights1))
78            yhat = tf.sigmoid(tf.matmul(y2, weights2))
79            yerror = tf.subtract(yhat, target)
80            loss1 = tf.nn.l2_loss(yerror)
81            L2_sqr = tf.nn.l2_loss(weights) + tf.nn.l2_loss(weights1) + tf.nn.l2_loss(weights2)
82            loss = loss1 + lambda_2 * L2_sqr
83            
84            input_test = tf.constant(bias_with_x_test)
85            target_test = tf.constant(np.transpose([y_test]).astype(np.float32))
86            y1_test = tf.nn.relu(tf.matmul(input_test, weights))
87            y2_test = tf.nn.relu(tf.matmul(y1_test, weights1))
88            yhat_test = tf.sigmoid(tf.matmul(y2_test, weights2))
89
90            update_weights = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)
91
92            for i in range(training_steps):
93                update_weights.run()
94                losses.append(loss.eval())
95            # Training is done, get the final values for the graphs
96            betas = weights.eval()
97            yhat = yhat.eval()
98            error = np.sqrt(((y_train-yhat)*(y_train-yhat)).mean()) #RMSEの算出
99
100            size = len(y_train)
101            MAE = (np.sum(np.abs(y_train-yhat)))/size
102                        
103            yhat_test = yhat_test.eval()
104            size = len(y_test)
105            MAE_test = (np.sum(np.abs(y_test-yhat_test)))/size
106                        
107            global MAE_train_ave
108            MAE_train_ave += MAE
109            
110            global MAE_test_ave
111            MAE_test_ave += MAE_test
112            
113    Train = MAE_train_ave/10.0        
114    Test = MAE_test_ave/10.0
115    print('学習誤差 : %s' % Train)
116    print('テスト誤差 : %s' % Test)
117   
118
119def main():
120    with open('../text/text2/label_text2.txt', 'r') as label:
121        a_list = [line.rstrip() for line in label]
122
123    for label in a_list:
124        process(label)
125        
126if __name__ == '__main__':
127    main()
128    
129