def network(companyname): seq_input, seq_output, keep_prob = define_placeholders(sequence_length, prediction_length) company = load_company_data(companyname) weights, biases = load_variables( variables_device, companyname) model_output = model(seq_input, weights, biases, keep_prob, companyname) cost = model_compilation(model_output, seq_output) init = tf.global_variables_initializer() run_model(init, model_output, companyname,\ company, seq_input, seq_output,\ keep_prob, dropout, cost, batch_size,\ prediction_length, sequence_length, processing_device)
from PIL import Image
import test
im = Image.open('drawing-1.png').convert('L')
im_array = np.asarray(im).ravel()
im_array = torch.from_numpy(im_array).type(torch.FloatTensor)
test.showImage(im_array)
im_array = im_array.unsqueeze(0)
out = test.model(im_array)
_, pred = torch.max(out.data, 1)
print(int(pred))
def predict_mask(input, threshold):
output = model(input.to(device))
output = torch.sigmoid(output).detach().cpu().numpy()
pred = output > threshold
return pred
def main():
# model config
dtype = tf.float32
model_name = 'model'
model_version = 1
len_X = 1032
len_Y = 2
len_XY = len_X + len_Y
# X/Y Placeholder and model
X = tf.placeholder(dtype=dtype, shape=(None, len_X))
Y = tf.placeholder(dtype=dtype, shape=(None, len_Y))
net, loss = model(x=X, y=Y, in_size=len_X, out_size=len_Y)
# session restore
sess = tf.Session()
saver = tf.train.Saver()
saver_path = './model_ckpt/'
last_ckpt_path = tf.train.latest_checkpoint(checkpoint_dir=saver_path)
saver.restore(sess=sess, save_path=last_ckpt_path)
# export builder
builder = tf.saved_model.builder.SavedModelBuilder('{:s}/{:d}'.format(model_name, model_version))
# tensor wrapping
regression_input = tf.saved_model.utils.build_tensor_info(X)
regression_output = tf.saved_model.utils.build_tensor_info(net)
# regression_loss = tf.saved_model.utils.build_tensor_info(loss)
# signature mapping
regression_signatures = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs=dict({
tf.saved_model.signature_constants.REGRESS_INPUTS:
regression_input
}),
outputs=dict({
tf.saved_model.signature_constants.REGRESS_OUTPUTS:
regression_output,
# 'loss': regression_loss
}),
method_name=
tf.saved_model.signature_constants.REGRESS_METHOD_NAME
)
)
# prediction signature (?)
tensor_info_x = tf.saved_model.utils.build_tensor_info(X)
tensor_info_y = tf.saved_model.utils.build_tensor_info(net)
prediction_signatures = tf.saved_model.signature_def_utils.build_signature_def(
inputs={'x': tensor_info_x},
outputs={'y': tensor_info_y},
method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME
)
legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op')
builder.add_meta_graph_and_variables(
sess=sess, tags=[tf.saved_model.tag_constants.SERVING],
signature_def_map=dict({
'predict': prediction_signatures,
'regression_value': regression_signatures
}),
legacy_init_op=legacy_init_op,
# main_op = tf.tables_initializer(),
strip_default_attrs=True,
)
builder.save()
# fin
return
im = cv2.dilate(edges,
cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5)),
iterations=2)
new_img = copy.deepcopy(im)
contours, hierarchy = cv2.findContours(new_img, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
#im = cv2.drawContours(im, contours, -1, (120,120,120), 6)
arr = []
i = 0
for cnt in contours:
#if cv2.contourArea(cnt)> 10000:
x, y, w, h = cv2.boundingRect(cnt)
arr.append((x, y, w, h))
model = test.model()
arr = sorted(arr, key=lambda x: x[0])
arrnew = []
for i in range(len(arr)):
k = 0
arrcpy = copy.deepcopy(arr)
del arrcpy[i]
for j in range(len(arrcpy)):
if arr[i][0] > arrcpy[j][0] and arr[i][1] > arrcpy[j][1] and (
arr[i][2] + arr[i][0]) < (arrcpy[j][2] + arrcpy[j][0]) and (
arr[i][3] + arr[i][0]) < (arrcpy[j][3] + arr[j][0]):
k += 1
if k == 0:
import nyt
import test
def merge(techs, arr):
output1 = {}
output2 = {}
# append to temp
for lst in arr:
for el in lst:
if str(el[0]) not in output1:
output1[str(el[0])] = 0
output1[str(el[0])] += el[1]
# match output1
for tech in techs:
output2[tech["technology_name"]] = output1[str(tech["technology_id"])]
return output2
out1 = nyt.model(technologies)
out2 = test.model(technologies)
print merge(technologies, [out1, out2])
from keras.optimizers import Adam
from keras.callbacks import ModelCheckpoint
from keras.utils import np_utils
import pandas as pd
import keras
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
from keras.utils import np_utils
import test as testing
test = pd.read_csv("emnist-balanced-test.csv", header=None)
x_test = test.iloc[:, 1:]
x_test = np.asarray(x_test)
model = testing.model()
model.load_weights("output/Weights.h5")
y_test = test.iloc[:, 0]
print np.asarray(y_test)
arr = []
correct = 0
for i in range(len(y_test)):
x = x_test[i]
x = x.reshape(1, 28, 28, 1).astype('float32')
x /= 255
model.predict(x)
out = model.predict(x)
arr.append(np.argmax(out))
print i
if arr[i] == y_test[i]: