17 KiB
17 KiB
In [1]:
import keras from keras.models import load_model from keras import utils import numpy as np import matplotlib.pyplot as plt import pandas as pd import math import tensorflow as tf # Importing matplotlib to plot images. import matplotlib.pyplot as plt import numpy as np %matplotlib inline # Importing SK-learn to calculate precision and recall import sklearn from sklearn import metrics from sklearn.model_selection import train_test_split, cross_val_score, LeaveOneGroupOut from sklearn.utils import shuffle # Used for graph export from tensorflow.python.framework import graph_util from tensorflow.python.framework import graph_io from keras import backend as K from keras import regularizers import pickle as pkl import h5py from pathlib import Path import os.path import sys import datetime import time target_names = ["Knuckle", "Finger"]
Using TensorFlow backend. /usr/local/lib/python3.6/dist-packages/requests/__init__.py:91: RequestsDependencyWarning: urllib3 (1.25.2) or chardet (3.0.4) doesn't match a supported version! RequestsDependencyWarning)
In [4]:
# the data, split between train and test sets df = pd.read_pickle("DataStudyCollection/df_blobs_area.pkl") lst = df.userID.unique() np.random.seed(42) np.random.shuffle(lst) test_ids = lst[-5:] train_ids = lst[:-5] print(train_ids, test_ids) print(len(train_ids), ":", len(test_ids)) print(len(train_ids) / len(lst), ":", len(test_ids)/ len(lst)) df = df[df.userID.isin(train_ids) | df.userID.isin(test_ids) & (df.Version == "Normal")] print(len(df))
[ 1 2 9 6 4 14 17 16 12 3 10 18 5] [13 8 11 15 7] 13 : 5 0.7222222222222222 : 0.2777777777777778 503886
In [3]:
x = np.vstack(df.Blobs) x = x.reshape(-1, 27, 15, 1)
In [4]:
# convert class vectors to binary class matrices (one-hot notation) num_classes = 2 y = utils.to_categorical(df.InputMethod, num_classes)
In [5]:
i = 1 plt.imshow(x[i].reshape(27, 15)) #np.sqrt(784) = 28 plt.title("Label for image %i is: %s" % (i, y[i]))
Out[5]:
Text(0.5, 1.0, 'Label for image 1 is: [1. 0.]')
In [6]:
# If GPU is not available: # GPU_USE = '/cpu:0' #config = tf.ConfigProto(device_count = {"GPU": 1}) # If GPU is available: config = tf.ConfigProto() config.log_device_placement = True config.allow_soft_placement = True config.gpu_options.allow_growth=True config.gpu_options.allocator_type = 'BFC' # Limit the maximum memory used config.gpu_options.per_process_gpu_memory_fraction = 0.4 # set session config tf.keras.backend.set_session(tf.Session(config=config))
In [7]:
loadpath = "./ModelSnapshots/CNN-33767.h5" model = load_model(loadpath)
In [8]:
%%time lst = [] batch = 100 for i in range(0, len(x), batch): _x = x[i: i+batch] lst.extend(model.predict(_x))
CPU times: user 1min 28s, sys: 9.52 s, total: 1min 37s Wall time: 1min
In [9]:
df["InputMethodPred"] = lst df.InputMethodPred = df.InputMethodPred.apply(lambda x: np.argmax(x))
In [10]:
df_train = df[df.userID.isin(train_ids)] df_test = df[df.userID.isin(test_ids) & (df.Version == "Normal")]
In [11]:
print(sklearn.metrics.confusion_matrix(df_train.InputMethod.values, df_train.InputMethodPred.values, labels=[0, 1])) cm = sklearn.metrics.confusion_matrix(df_train.InputMethod.values, df_train.InputMethodPred.values, labels=[0, 1]) cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis] print(cm) print("Accuray: %.3f" % sklearn.metrics.accuracy_score(df_train.InputMethod.values, df_train.InputMethodPred.values)) print("Recall: %.3f" % metrics.recall_score(df_train.InputMethod.values, df_train.InputMethodPred.values, average="macro")) print("Precision: %.3f" % metrics.average_precision_score(df_train.InputMethod.values, df_train.InputMethodPred.values, average="macro")) #print("F1-Score: %.3f" % metrics.f1_score(df_train.InputMethod.values, df_train.InputMethodPred.values, average="macro")) print(sklearn.metrics.classification_report(df_train.InputMethod.values, df_train.InputMethodPred.values, target_names=target_names))
[[124207 12765]
[ 6596 322276]]
[[0.90680577 0.09319423]
[0.02005644 0.97994356]]
Accuray: 0.958
Recall: 0.943
Precision: 0.957
precision recall f1-score support
Knuckle 0.95 0.91 0.93 136972
Finger 0.96 0.98 0.97 328872
micro avg 0.96 0.96 0.96 465844
macro avg 0.96 0.94 0.95 465844
weighted avg 0.96 0.96 0.96 465844
In [12]:
print(sklearn.metrics.confusion_matrix(df_test.InputMethod.values, df_test.InputMethodPred.values, labels=[0, 1])) cm = sklearn.metrics.confusion_matrix(df_test.InputMethod.values, df_test.InputMethodPred.values, labels=[0, 1]) cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis] print(cm) print("Accuray: %.3f" % sklearn.metrics.accuracy_score(df_test.InputMethod.values, df_test.InputMethodPred.values)) print("Recall: %.3f" % metrics.recall_score(df_test.InputMethod.values, df_test.InputMethodPred.values, average="macro")) print("Precision: %.3f" % metrics.average_precision_score(df_test.InputMethod.values, df_test.InputMethodPred.values, average="macro")) #print("F1-Score: %.3f" % metrics.f1_score(df_test.InputMethod.values, df_test.InputMethodPred.values, average="macro")) print(sklearn.metrics.classification_report(df_test.InputMethod.values, df_test.InputMethodPred.values, target_names=target_names))
[[ 8384 1037]
[ 1028 27593]]
[[0.88992676 0.11007324]
[0.03591768 0.96408232]]
Accuray: 0.946
Recall: 0.927
Precision: 0.956
precision recall f1-score support
Knuckle 0.89 0.89 0.89 9421
Finger 0.96 0.96 0.96 28621
micro avg 0.95 0.95 0.95 38042
macro avg 0.93 0.93 0.93 38042
weighted avg 0.95 0.95 0.95 38042
In [ ]:
Export¶
In [13]:
output_node_prefix = "output_node" num_output = 1 pred = [None]*num_output pred_node_names = [None]*num_output for i in range(num_output): pred_node_names[i] = output_node_prefix+str(i) pred[i] = tf.identity(model.outputs[i], name=pred_node_names[i]) print('output nodes names are: ', pred_node_names) output_node_prefix = pred_node_names[0]
output nodes names are: ['output_node0']
In [14]:
model.inputs
Out[14]:
[<tf.Tensor 'conv2d_1_input:0' shape=(?, 27, 15, 1) dtype=float32>]
In [15]:
sess = K.get_session()
In [16]:
output_path = "./Models/" output_file = "CNN.pb"
In [17]:
from tensorflow.python.framework import graph_util from tensorflow.python.framework import graph_io constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph.as_graph_def(), pred_node_names) graph_io.write_graph(constant_graph, output_path, output_file, as_text=False) print('Saved the freezed graph at: ', (output_path + output_file))
INFO:tensorflow:Froze 30 variables. INFO:tensorflow:Converted 30 variables to const ops. Saved the freezed graph at: ./Models/CNN.pb