def get_preds(data_set, model_type = None, model_path = SAVE_MODEL_DIR, output_file_name='./data/last_preds.json'): pred_handler = PredictionHandler() data = DataHolder(data_set) with tf.Graph().as_default(): start = time.time() if model_type == 'l2_attn': model = l2_attn.TFModel(EmbeddingHolder().get_embeddings_mat(), True) elif model_type == 'l2_2attn': model = l2_2attn.TFModel(EmbeddingHolder().get_embeddings_mat(), True) elif model_type == 'l2_double_2attn': model = l2_double_2attn.TFModel(EmbeddingHolder().get_embeddings_mat(), True) elif model_type == 'ce_attn': model = ce_attn.TFModel(EmbeddingHolder().get_embeddings_mat(), True) elif model_type == 'ce_2attn': model = ce_2attn.TFModel(EmbeddingHolder().get_embeddings_mat(), True) else: return print "\nRebuild graph took " + str(time.time() - start) + " seconds" saver = tf.train.Saver() with tf.Session() as session: session = session saver.restore(session, model_path) print 'Restored model. Predicting....' preds = model.predict(session, data) pred_handler.write_preds(preds, output_file_name) print 'Done Predicting'
def get_classifier_results(data_set, model_path = SAVE_MODEL_DIR):
data = DataHolder(data_set)
with tf.Graph().as_default():
start = time.time()
model = PassClassifier(EmbeddingHolder().get_embeddings_mat(), True)
print "\nRebuild graph took " + str(time.time() - start) + " seconds"
saver = tf.train.Saver()
with tf.Session() as session:
session = session
saver.restore(session, model_path)
print 'Restored model. Predicting....'
preds = model.predict(session, data)
list_preds = list()
for batch in preds:
for row in batch:
list_preds.append(row)
preds = np.asarray(list_preds)
y = data.get_full_selected()
f = open('classifier_check_log', 'w')
f.write("File to check results of classifier:\n Preds Y")
for p_val, y_val in zip(preds, y):
f.write(str(p_val) + ' ' + str(y_val))
f.write('\n')
classifier_eval(preds, y)
print 'Done Predicting'
def keep_training(model_path = SAVE_MODEL_DIR, prev_best_rouge=0):
pred_handler = PredictionHandler()
data = DataHolder('train')
with tf.Graph().as_default():
start = time.time()
model = l2_double_2attn.TFModel(EmbeddingHolder().get_embeddings_mat())
print "\nRebuild graph took " + str(time.time() - start) + " seconds"
saver = tf.train.Saver()
with tf.Session() as session:
session = session
saver.restore(session, model_path)
print 'Restored model. Predicting....'
merged = tf.summary.merge_all()
model.best_rouge = prev_best_rouge
losses = model.fit(session, saver, merged, data)
model.train_writer.close()
model.test_writer.close()
model.log.close()
print 'Done continueing'
def build_model():
model = Sequential()
embedding_matrix = EmbeddingHolder().get_embeddings_mat()
# BUILD ACTUAL MODEL
embedding_layer = Embedding(
228999,
EMBEDDING_DIM,
weights=[embedding_matrix],
input_length=INPUT_MAX_LENGTH,
mask_zero=True
)
s2s_layer = Seq2Seq(
batch_input_shape=(None, INPUT_MAX_LENGTH, EMBEDDING_DIM),
hidden_dim=HIDDEN_DIM,
output_length=OUTPUT_MAX_LENGTH,
output_dim=228999,
depth=DEPTH
)
# ADD UP ACTUAL MODEL
model.add(embedding_layer)
model.add(s2s_layer)
model.compile(loss='mse', optimizer='adam', metrics=['accuracy', 'precision', 'recall'])
print 'Model Built'
return model
def build(self):
self.add_placeholders()
self.pred = self.add_prediction_op()
self.loss = self.add_loss_op(self.pred)
self.train_op = self.add_training_op(self.loss)
def __init__(self, embeddings):
self.pretrained_embeddings = embeddings
self.log = open(LOG_FILE_DIR, "a")
self.build()
if __name__ == "__main__":
print 'Starting, and now printing to log.txt'
data = TFDataHolder('train')
embeddings = EmbeddingHolder().get_embeddings_mat()
with tf.Graph().as_default():
start = time.time()
model = TFModel(embeddings)
model.log.write("\nBuild graph took " + str(time.time() - start) + " seconds")
init = tf.global_variables_initializer()
saver = tf.train.Saver()
model.log.write('\ninitialzed variables')
config = tf.ConfigProto()
# config.gpu_options.allow_growth=True
# config.gpu_options.per_process_gpu_memory_fraction = 0.6
with tf.Session(config=config) as session:
session.run(init)
model.log.write('\nran init, fitting.....')
losses = model.fit(session, saver, data)