def save_models_to_disk_static(top_model, root, new_name):
top_model.model = lm(root + '{}.h5'.format(new_name))
top_model.inf_encoder_model = lm(root +
'{}_inf_encoder.h5'.format(new_name))
top_model.inf_intent_classifier = lm(
root + '{}_inf_intent.h5'.format(new_name))
top_model.inf_decoder_model = lm(root +
'{}_inf_decoder.h5'.format(new_name))
def load_models_from_disk(self, root, new_name):
# self.model = lm(root + '{}.h5'.format(new_name))
self.inf_encoder_model = lm(root +
'{}_inf_encoder.h5'.format(new_name))
self.inf_intent_classifier = lm(root +
'{}_inf_intent.h5'.format(new_name))
self.inf_decoder_model = lm(root +
'{}_inf_decoder.h5'.format(new_name))
def load_model(model_path):
"""Loads the model using a json description of the model and a hdf file containing the weights.
Args:
model_path: path to the model.h5 file.
Returns:
model: model object
"""
return lm(model_path)
from PIL import Image as im
def open(p: str) -> np.ndarray:
img = im.open(os.path.join(path, p)).convert('L')
(width, height) = img.size
if width > 28 and height > 28:
img = img.resize((28, 28), im.ANTIALIAS)
else:
pass
(width, height) = img.size
img = list(img.getdata())
img = np.array(img)
img = img.reshape((height, width))
return img
def norm(x: np.ndarray):
ret = x / 255
ret = np.expand_dims(ret, axis=0)
ret = np.expand_dims(ret, axis=3)
return ret
model = lm('IRC_MNIST.h5')
img = open('MyTestNum1.png')
img = norm(img)
res = model.predict(img)
print(f"Розпізнана цифра: {np.argmax(res)}")
def load_model_2(self, path):
# self.sess = tf.Session()
# with self.sess.as_default():
# # backend.set_session(self.sess)
with CustomObjectScope({'GlorotUniform': glorot_uniform()}):
return lm(path)
# frame['eventid'] = dataclasses.I3VectorDouble(eventid)
# frame['TCN'] = dataclasses.I3VectorDouble(prediction_vector)
# frame['truth'] = dataclasses.I3VectorDouble(truth_vector)
def savefeatures(frame, fin=None, names=None):
for i, name in enumerate(names):
fin[i].append(frame[name])
use_truth = 1
#Tensorflow stuff
if use_truth == 0:
from tensorflow.keras.models import load_model as lm
model = lm(
'/home/sstray/test/condor/corsikafiles/corsika_model_test/classification_model'
)
#
import os
from I3Tray import *
from icecube import dataio
from icecube import dataclasses
from icecube import simclasses
from icecube import recclasses
import numpy as np
#import cProfile, pstats, StringIO
import argparse
import matplotlib.path as mpath
bordercoords = np.array([(-256.1400146484375, -521.0800170898438),
(-132.8000030517578, -501.45001220703125),
(-9.13000011444092, -481.739990234375),
parser.add_argument('-rad','--radius')
parser.add_argument('-th','--threshold')
args = parser.parse_args()
if args.radius:
rad = args.radius
else:
rad = 75
print('No radius supplied. Will default to 75')
if args.threshold:
th = args.threshold
else:
th = 0.9
print('No threshold supplied. Will default to 0.9')
meanstd = np.loadtxt('/home/sstray/test/condor/singu/meanstd.txt')
model = lm('/home/sstray/test/condor/singu/my_model/classification_model')
def get_groups(features_in,event_in):
features = features_in
eventnum = event_in
grouped_features = ([features[np.isin(eventnum,[i])] for i in np.unique(eventnum)])
grouped_features = [np.vstack((np.array(i),np.zeros((100,features.shape[1]))))[:100] for i in grouped_features if len(i)<100 if len(i)>10]
grouped_features = np.dstack(grouped_features)
grouped_features = np.rollaxis(grouped_features,2,0)
return grouped_features
filelist = args.infiles.split(',')
print(filelist)
for nums in range(len(filelist)):
hf = h5py.File(filelist[nums],'r')