Deep Learning p8 ValueError when printing accuracy.

by: mtbrands, 16 days ago

I'm following the DeepLearning tutorials and they have been great so far! A couple of things have become outdated but I have found updates for these. One error I found however is not really clear to me.
Traceback (most recent call last):
  File "-/TFDeepLearning/", line 135, in <module>
  File "-/", line 132, in test_neural_network
  File "", line 570, in eval
    return _eval_using_default_session(self, feed_dict, self.graph, session)
  File "", line 4455, in _eval_using_default_session
    return, feed_dict)
  File "", line 889, in run
  File "", line 1089, in _run
    np_val = np.asarray(subfeed_val, dtype=subfeed_dtype)
  File "", line 531, in asarray
    return array(a, dtype, copy=False, order=order)
ValueError: setting an array element with a sequence.

I've narrowed the source down to the feed_dict . But comparing the code to the video tutorials and code provided on this site I've been unable to find a solution.

import tensorflow as tf
import pickle
import numpy as np
import nltk
from nltk.tokenize import word_tokenize
from nltk.stem import WordNetLemmatizer
lemmatizer = WordNetLemmatizer()

n_nodes_hl1 = 500
n_nodes_hl2 = 500

n_classes = 2

batch_size = 32
total_batches = int(1600000/batch_size)
hm_epochs = 10

x = tf.placeholder('float')
y = tf.placeholder('float')

hidden_1_layer = {'f_fum':n_nodes_hl1,
                  'weight':tf.Variable(tf.random_normal([84, n_nodes_hl1])),

hidden_2_layer = {'f_fum':n_nodes_hl2,
                  'weight':tf.Variable(tf.random_normal([n_nodes_hl1, n_nodes_hl2])),

output_layer = {'f_fum':None,
                'weight':tf.Variable(tf.random_normal([n_nodes_hl2, n_classes])),

def neural_network_model(data):
    l1 = tf.add(tf.matmul(data,hidden_1_layer['weight']), hidden_1_layer['bias'])
    l1 = tf.nn.relu(l1)
    l2 = tf.add(tf.matmul(l1,hidden_2_layer['weight']), hidden_2_layer['bias'])
    l2 = tf.nn.relu(l2)
    output = tf.matmul(l2,output_layer['weight']) + output_layer['bias']
    return output

saver = tf.train.Saver()
#saver = tf.train.import_meta_graph('./model.ckpt.meta')
tf_log = 'tf.log'

def train_neural_network(x):
    prediction = neural_network_model(x)
    cost = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(logits=prediction,labels=y) )
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(cost)
    with tf.Session() as sess:
            epoch = int(open(tf_log,'r').read().split('n')[-2])+1
            epoch = 1

        while epoch <= hm_epochs:
            if epoch != 1:
            epoch_loss = 1
            with open('lexicon-2500-2638.pickle','rb') as f:
                lexicon = pickle.load(f)
                print('lexicon length:',len(lexicon))
            with open('train_set_shuffled.csv', buffering=20000, encoding='latin-1') as f:
                batch_x = []
                batch_y = []
                batches_run = 0
                for line in f:
                    label = line.split(':::')[0]
                    tweet = line.split(':::')[1]
                    current_words = word_tokenize(tweet.lower())
                    current_words = [lemmatizer.lemmatize(i) for i in current_words]

                    features = np.zeros(len(lexicon))

                    for word in current_words:
                        if word.lower() in lexicon:
                            index_value = lexicon.index(word.lower())
                            # OR DO +=1, test both
                            features[index_value] += 1
                    line_x = list(features)
                    line_y = eval(label)
                    if len(batch_x) >= batch_size:
#possible problem source in feed)_dict
                        _, c =[optimizer, cost], feed_dict={x: np.array(batch_x),
                                                                  y: np.array(batch_y)})
                        epoch_loss += c
                        batch_x = []
                        batch_y = []
                        batches_run +=1
                        print('Batch run:',batches_run,'/',total_batches,'| Epoch:',epoch,'| Batch Loss:',c,)

  , "./model.ckpt")
            print('Epoch', epoch, 'completed out of',hm_epochs,'loss:',epoch_loss)
            with open(tf_log,'a') as f:
            epoch +=1


def test_neural_network():
    prediction = neural_network_model(x)
    with tf.Session() as sess:
        for epoch in range(hm_epochs):
            except Exception as e:
            epoch_loss = 0
        correct = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))
        accuracy = tf.reduce_mean(tf.cast(correct, 'float'))
        #accuracy = tf.cast(correct, 'float')
        feature_sets = []
        labels = []
        counter = 0
        with open('processed-test-set.csv', buffering=20000) as f:
            for line in f:
                    features = list(eval(line.split('::')[0]))
                    label = list(eval(line.split('::')[1]))
                    counter += 1
        #possible problem source
        test_x = np.array(feature_sets)
        test_y = np.array(labels)
       #this line gets a valuerror.


I had to make some adjustments for it to work to where it does, most importantly the amount of words in the lexicon.
Is this a common problem? What process should I go through to find a correct solution?

You must be logged in to post. Please login or register an account.