Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit 5c54dc5

Browse filesBrowse files
ozgesevgiliozgesevgili
committed
Add dropout.
1 parent 256c693 commit 5c54dc5
Copy full SHA for 5c54dc5

File tree

Expand file treeCollapse file tree

1 file changed

+37
-31
lines changed
Open diff view settings
Filter options
Expand file treeCollapse file tree

1 file changed

+37
-31
lines changed
Open diff view settings
Collapse file

‎supervised/nn.py‎

Copy file name to clipboardExpand all lines: supervised/nn.py
+37-31Lines changed: 37 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -128,9 +128,9 @@ def get_parameters():
128128
'graph embeddings and doc2vec.')
129129

130130
parser.add_argument('-training_corpus',
131-
help="Path to a training corpus contaning both positive and negative examples"
132-
" in the format -- positive sample, starting location of the ambigous word, "
133-
"endinglocation of the ambigous word , true_url, context, negative_samples_urls -- "
131+
help="Path for a training corpus containing both positive and negative examples"
132+
" in the format -- positive sample, starting location of the ambiguous word, "
133+
"ending location of the ambiguous word , true_url, context, negative_samples_urls -- "
134134
"seperated by the tabs in text form.",
135135
default='/Users/sevgili/Ozge-PhD/DBpedia-datasets/training-datasets/csv/negative_samples_filtered_randomly_3.tsv')
136136

@@ -151,8 +151,8 @@ def get_parameters():
151151
"and values contain the long abstracts of each entity.",
152152
default='/Users/sevgili/Ozge-PhD/DBpedia-datasets/outputs/databases/long_abstracts.db')
153153

154-
parser.add_argument('-gpu', help="The decision of cpu and gpu, if you would like to use gpu please give only number, e.g. -gpu 0"
155-
"(default cpu).",
154+
parser.add_argument('-gpu', help="The decision of cpu and gpu, if you would like to use gpu please "
155+
"give only number, e.g. -gpu 0 (default cpu).",
156156
default='/cpu')
157157

158158
parser.add_argument('-learning_rate', help="Learning rate for the optimizer in neural network (default 0.005).",
@@ -189,6 +189,9 @@ def get_parameters():
189189
parser.add_argument('-l2_beta', help="The beta parameter of L2 loss (default 0, means no L2 loss!)",
190190
default=0.0, type=float)
191191

192+
parser.add_argument('-dropout', help="The keep probability of dropout (default 1.0, means dropout!)",
193+
default=1.0, type=float)
194+
192195
args = parser.parse_args()
193196

194197
device = args.gpu
@@ -207,7 +210,7 @@ def get_parameters():
207210

208211
return args.training_corpus, args.graph_embedding, args.doc2vec, args.lookup_db, args.long_abstracts_db, \
209212
args.learning_rate, args.training_epochs, args.h1_size, args.h2_size, h3_size, h4_size, h5_size,\
210-
args.h6_size, device, args.l2_beta
213+
args.h6_size, device, args.l2_beta, args.dropout
211214

212215

213216
def forward_propagation(x_size, y_size, h1_size, h2_size, h3_size, h4_size, h5_size, h6_size, l2_beta):
@@ -232,12 +235,12 @@ def forward_propagation(x_size, y_size, h1_size, h2_size, h3_size, h4_size, h5_s
232235
}
233236

234237
# Forward propagation
235-
h1 = tf.add(tf.matmul(X, weights['w1']), biases['b1'])
236-
h2 = tf.add(tf.matmul(h1, weights['w2']), biases['b2'])
237-
h3 = tf.add(tf.matmul(h2, weights['w3']), biases['b3'])
238-
h4 = tf.add(tf.matmul(h3, weights['w4']), biases['b4'])
239-
h5 = tf.add(tf.matmul(h4, weights['w5']), biases['b5'])
240-
h6 = tf.add(tf.matmul(h5, weights['w6']), biases['b6'])
238+
h1 = tf.nn.dropout(tf.add(tf.matmul(X, weights['w1']), biases['b1']), keep_prob)
239+
h2 = tf.nn.dropout(tf.add(tf.matmul(h1, weights['w2']), biases['b2']), keep_prob)
240+
h3 = tf.nn.dropout(tf.add(tf.matmul(h2, weights['w3']), biases['b3']), keep_prob)
241+
h4 = tf.nn.dropout(tf.add(tf.matmul(h3, weights['w4']), biases['b4']), keep_prob)
242+
h5 = tf.nn.dropout(tf.add(tf.matmul(h4, weights['w5']), biases['b5']), keep_prob)
243+
h6 = tf.nn.dropout(tf.add(tf.matmul(h5, weights['w6']), biases['b6']), keep_prob)
241244

242245
# if h6 is specified
243246
if h6_size:
@@ -286,17 +289,18 @@ def forward_propagation(x_size, y_size, h1_size, h2_size, h3_size, h4_size, h5_s
286289
return yhat, l2_loss
287290

288291

289-
def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs):
292+
def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs, keep_probability):
290293
sess.run(tf.global_variables_initializer())
291294

292295
for epoch in range(training_epochs):
293-
_, current_loss = sess.run([optimizer, loss], feed_dict={X: train_inputs, y: train_outputs})
296+
_, current_loss = sess.run([optimizer, loss], feed_dict={X: train_inputs,
297+
y: train_outputs, keep_prob:keep_probability})
294298

295299

296300
if __name__ == "__main__":
297301
training_corpus, path_graphembed, path_doc2vec, path_lookupdb, path_longabsdb, \
298302
learning_rate, training_epochs, h1_size, h2_size, h3_size, h4_size, h5_size, h6_size, \
299-
device, l2_beta = get_parameters()
303+
device, l2_beta, dropout = get_parameters()
300304

301305
print('hidden sizes:', h1_size, h2_size, h3_size, h4_size, h5_size, h6_size)
302306

@@ -314,6 +318,7 @@ def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs):
314318
# Symbols
315319
X = tf.placeholder(tf.float32, shape=[None, x_size])
316320
y = tf.placeholder(tf.float32, shape=[None, y_size])
321+
keep_prob = tf.placeholder(tf.float32)
317322

318323
# Forward propagation
319324
yhat, l2_loss = forward_propagation(x_size, y_size, h1_size, h2_size, h3_size, h4_size, h5_size, h6_size, l2_beta)
@@ -370,7 +375,7 @@ def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs):
370375
val_outputs = outputs[val_id]
371376

372377
train(sess=sess, optimizer=optimizer, loss=loss, train_inputs=train_inputs,
373-
train_outputs=train_outputs, training_epochs=training_epochs)
378+
train_outputs=train_outputs, training_epochs=training_epochs, keep_probability=dropout)
374379

375380
'''
376381
for epoch in range(training_epochs):
@@ -385,10 +390,10 @@ def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs):
385390
summary_str = sess.run(train_summary_op, feed_dict={X: train_inputs, y: train_outputs})
386391
summary_writer.add_summary(summary_str, epoch)
387392
'''
388-
cv_accuracy, cv_precision, cv_recall, cv_f1 = accuracy.eval({X: val_inputs, y: val_outputs}), \
389-
precision.eval({X: val_inputs, y: val_outputs}), \
390-
recall.eval({X: val_inputs, y: val_outputs}), \
391-
f1.eval({X: val_inputs, y: val_outputs})
393+
cv_accuracy, cv_precision, cv_recall, cv_f1 = accuracy.eval({X: val_inputs, y: val_outputs, keep_prob:1}), \
394+
precision.eval({X: val_inputs, y: val_outputs, keep_prob:1}), \
395+
recall.eval({X: val_inputs, y: val_outputs, keep_prob:1}), \
396+
f1.eval({X: val_inputs, y: val_outputs, keep_prob:1})
392397
cv_accuracies.append(cv_accuracy)
393398
cv_precisions.append(cv_precision)
394399
cv_recalls.append(cv_recall)
@@ -401,24 +406,25 @@ def train(sess, optimizer, loss, train_inputs, train_outputs, training_epochs):
401406

402407
# Train model
403408
train(sess=sess, optimizer=optimizer, loss=loss, train_inputs=inputs,
404-
train_outputs=outputs, training_epochs=training_epochs)
409+
train_outputs=outputs, training_epochs=training_epochs, keep_probability=dropout)
405410

406411
# Test model
407412
pred = tf.nn.sigmoid(yhat)
408413
correct_prediction = tf.equal(tf.round(pred), y)
409414

410415
print("Training Report", classification_report(y_true=outputs,
411-
y_pred=tf.round(pred).eval({X: inputs, y: outputs})))
416+
y_pred=tf.round(pred).eval({X: inputs, y: outputs, keep_prob:dropout})))
412417

413418
print("Test Report", classification_report(y_true=test_outputs,
414-
y_pred=tf.round(pred).eval({X: test_inputs, y: test_outputs})))
419+
y_pred=tf.round(pred).eval({X: test_inputs, y: test_outputs, keep_prob:1})))
415420

416421
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
417-
print("Training Accuracy:", accuracy.eval({X: inputs, y: outputs}))
418-
print("Training Precision:", precision.eval({X: inputs, y: outputs}),
419-
"Training Recall:", recall.eval({X: inputs, y: outputs}), "Training F1:", f1.eval({X: inputs, y: outputs}))
420-
421-
print("Test Accuracy:", accuracy.eval({X: test_inputs, y: test_outputs}))
422-
print("Test Precision:", precision.eval({X: test_inputs, y: test_outputs}),
423-
"Test Recall:", recall.eval({X: test_inputs, y: test_outputs}), " Test F1:",
424-
f1.eval({X: test_inputs, y: test_outputs}))
422+
print("Training Accuracy:", accuracy.eval({X: inputs, y: outputs, keep_prob:dropout}))
423+
print("Training Precision:", precision.eval({X: inputs, y: outputs, keep_prob:dropout}),
424+
"Training Recall:", recall.eval({X: inputs, y: outputs, keep_prob:dropout}),
425+
"Training F1:", f1.eval({X: inputs, y: outputs, keep_prob:dropout}))
426+
427+
print("Test Accuracy:", accuracy.eval({X: test_inputs, y: test_outputs, keep_prob:1}))
428+
print("Test Precision:", precision.eval({X: test_inputs, y: test_outputs, keep_prob:1}),
429+
"Test Recall:", recall.eval({X: test_inputs, y: test_outputs, keep_prob:1}), " Test F1:",
430+
f1.eval({X: test_inputs, y: test_outputs, keep_prob:1}))

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.