def find_similar(self, data, num_similar=5, return_distance=False): data = format_data(data) check_data(data) X = self.transform(data) dist, ind = self.knn.kneighbors(X=X, n_neighbors=num_similar, return_distance=True) if return_distance: return ind, dist return ind
def fit_transform(self, data, patient_ids=None):
data = format_data(data)
check_data(data)
X = pad(data)
X = self.model.predict(X)
X = X.reshape(-1, X.shape[-1])
if patient_ids is None:
patient_ids = np.arange(len(data))
self.patient_ids = {i: patient_ids[patientint(i / SEQ_LENGTH)] for i in range(len(X))}
self.patient_offsets = {i: patient_ids[patientint(i % SEQ_LENGTH)] for i in range(len(X))}
self._fit_X = X
self._fit_pca_X = self.pca.fit_transform(X)
self.knn.fit(self._fit_pca_X)
if args.freq:
if not args.load or not model:
print(
"No model loaded. Please load a model with --load or -l before doing frequency analysis."
)
exit()
magnitudes = []
for i in range(3, 6):
for j in range(1, 10):
frequency = j * 10**i
print("Frequency: " + str(j) + "x10^" + str(i))
data_obj_test = create_freq_data(frequency)
(x_test, y_test) = format_data(data_obj_test, args.timesteps)
magnitudes.append(
compute_magnitude(model, x_test, data_obj_test,
args.batch_size, frequency, model_name,
args.plot))
print("\n")
print("-- Low Pass Magnitudes --")
for mag in magnitudes:
print(mag[0])
print("\n")
print("-- Network Magnitudes --")
for mag in magnitudes:
print(mag[1])
from model import load_model
from data import create_data, format_data
import matplotlib.pyplot as plt
import numpy as np
data_obj_test = create_data("OOK", "LPF")
(x_test, y_test) = format_data(data_obj_test, 32)
fig, axs = plt.subplots(8, 1)
model = load_model("CNN3232")
output = model.predict(x_test, batch_size=32)
axs[0].plot(output, label="CNN3232")
plt.show()
X = self.model.predict(X)
X = X.reshape(-1, X.shape[-1])
if patient_ids is None:
patient_ids = np.arange(len(data))
self.patient_ids = {i: patient_ids[patientint(i / SEQ_LENGTH)] for i in range(len(X))}
self.patient_offsets = {i: patient_ids[patientint(i % SEQ_LENGTH)] for i in range(len(X))}
self._fit_X = X
self._fit_pca_X = self.pca.fit_transform(X)
self.knn.fit(self._fit_pca_X)
def transform(self, data)
data = format_data(data)
check_data(data)
assert self.pca is not None and self.knn is not None, 'Must call function "fit_transform" before using "transform"'
ends = np.clip([len(array) for array in data], 0, SEQ_LENGTH - 1)
X = pad(data)
X = self.model.predict(X)
X = X[np.arange(len(X)), ends]
return self.pca.transform(X)
def find_similar(self, data, num_similar=5, return_distance=False):
data = format_data(data)
check_data(data)
def generate_graphs(weighing_method='RS'):
''' Generate all the graphs and associated layout
based on the provided weighing method, RS if none is provided
'''
# Try if the default files are located in the root dir, otherwise initiate with example data
try:
csv_data = pd.read_csv('input_data.csv')
with open('input_data_labels.json') as input_data_labels_file:
input_data_labels = json.load(input_data_labels_file)
csv_labels = input_data_labels['csv_labels']
csv_type = input_data_labels['csv_type']
attributes = format_data(csv_data, csv_labels, csv_type,
weighing_method)
except:
attributes = format_data()
# Generate pie char for weights with header
weights_labels = [k for k, v in attributes.items()]
weights_values = [v['weight'] for k, v in attributes.items()]
weights_fig = go.Figure(data=[
go.Pie(labels=weights_labels,
values=weights_values,
textinfo='label+percent',
insidetextorientation='radial')
],
layout_margin=dict(l=50, r=50, t=50, b=50),
layout_showlegend=False)
# weights_fig.write_image('images/Ranks.eps')
weights_graph = [
html.Div(children=[
html.H1(children='Weights', style={'textAlign': 'center'}),
dcc.Dropdown(options=[{
'label': 'RS - Rank Sum',
'value': 'RS'
}, {
'label': 'ROC - Rank Order Centroid',
'value': 'ROC'
}],
value=weighing_method,
id='weights-selector'),
dcc.Graph(figure=weights_fig, id='weights-graph')
])
]
# Generate Dash graphs from figures for each attribute, with title in a <div>
attribute_graphs = [
graph_element(key, attribute) for key, attribute in attributes.items()
]
# Generate overall title and attributes title
title = [
html.H1(children='SMARTER Preference Elicitation Analysis Tool',
style={'textAlign': 'center'}),
html.Hr()
]
title_attributes = [
html.Hr(),
html.H1(children='Attributes', style={'textAlign': 'center'})
]
graphs = title + weights_graph + title_attributes + attribute_graphs
return graphs