TypeError: Expected binary or unicode string, got list Tensorflow

Question

Alright, so bear with me. So this is my code:

import tensorflow as tf
import os
import pickle
from numpy import asarray, reshape

os.chdir('PATH')

with open('xSensor.pkl','rb') as file:
    x_train = asarray(pickle.load(file))

with open('ySensor.pkl','rb') as file:
    y_train = asarray(pickle.load(file))

def neural_network(data):
    n_nodes_h1 = 1000
    n_nodes_h2 = 1000
    n_nodes_h3 = 500

    hidden_layer_1 = {
        'weights': tf.Variable(tf.random_normal([13, n_nodes_h1],dtype=tf.float64)),
        'biases': tf.Variable(tf.random_normal([n_nodes_h1],dtype=tf.float64))
    }

    #Omitting some code that just defines a couple more layers in the same format as above

    layer_1 = tf.matmul(data, hidden_layer_1['weights']) + hidden_layer_1['biases']
    layer_1 = tf.nn.relu(layer_1)

    #Omitting more code.

    output = tf.matmul(layer_3, output['weights']) + output['biases']

    return output


def train_network(x_t,y_t):
    x = tf.placeholder(tf.float64, shape=[None, 13])
    y = tf.placeholder(tf.float64)
    prediction = neural_network(x_t)
    y_t = reshape(y_t,(700,1))
    cost = tf.reduce_mean(tf.losses.mean_squared_error(labels=y_t, predictions=prediction))
    optimizer = tf.train.AdamOptimizer(0.005).minimize(cost) #learning rate by default is 0.01
    n_epochs = 1000

    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for _ in range(0, n_epochs):
            x_ = sess.run([optimizer,cost], feed_dict={x: x_t, y: y_t})
            print("Loss is: ", x_[1])



train_network(x_train,y_train)

And here's the error log:

Traceback (most recent call last):
  File "C:/Users/my system/Desktop/height_sensor.py", line 94, in <module>
    train_network(x_train,y_train)
  File "C:/Users/my system/Desktop/height_sensor.py", line 77, in train_network
    prediction = neural_network(x_t)
  File "C:/Users/my system/Desktop/height_sensor.py", line 59, in neural_network
    layer_1 = tf.matmul(data, hidden_layer_1['weights']) + hidden_layer_1['biases']
  File "C:\Users\my system\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\ops\math_ops.py", line 1844, in matmul
    a = ops.convert_to_tensor(a, name="a")
  ...etc...
TypeError: Expected binary or unicode string, got [15.0126, 1.38684, 27.6, 1.6323, -0.624113, 8.97763, 2.06581, 8.88303, -0.689839, 9.13284, 353.183, 349.178, 210.498]

Process finished with exit code 1

Sorry for posting so much code, I wanted to post waaaay less but I was worried that I might omit that one thing that was causing the error. If anyone could so much as point me in the right direction I'd be more than happy. Thanks

Answer 1

It's actually a good idea to post all your code, and especially all of your error traceback. I'm able to get your code to run with some fake data. My best guess is that there's something off about the way pickle is returning the value for x_train; possibly the array has too much nesting. I might be able to help more with the entire traceback, but at any rate I'd recommend outputting some of x_train and seeing if it's in the right format.