def GenerateHighPredictionGraph(net, data, config, directory, fname):
global reportinfo
# region Measure Prediction
best = 0
secondbest = 0
xaxis = ""
yaxis = ""
for inputa in net.inputs:
if inputa in reportinfo.usedpredictability:
continue
score = 0
for item in data["data"]:
net.reset()
for key in item:
if key == inputa:
net.setNode(key, m.unitscale(data["inputs"][key]["min"], item[key], data["inputs"][key]["max"]))
elif key in net.inputs:
net.setNode(key, 0)
net.process()
for out in net.outputs:
val = item[out]
if config["test-mode"] != "SimplePosi":
val = net.scale(out, item[out])
score += ge.Test_Output(net.getOutput()[out], val, config["test-mode"])
if score > best:
yaxis = xaxis
xaxis = inputa
secondbest = best
best = score
elif score > secondbest:
yaxis = inputa
secondbest = score
if yaxis == "":
return "<p>Not enough variables for graph.</p>"
# endregion
#region Handle graph
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
if not config["add-data"]:
MLGraphing.GraphNet(net, xaxis, yaxis, plt)
else:
MLGraphing.GraphNetData(net, data, xaxis, yaxis, net.classifier_output, 0, config["clump"], config["percents"], config["color"], plt)
if config["customize"]:
pyplot.show()
fig.savefig(directory + '/' + fname, bbox_inches='tight')
pyplot.close(fig)
out = GenerateGraphHTML(net, config["header"], fname, xaxis, yaxis, config["add-data"], config["color"],
config["clump"], config["percents"], reportinfo.usedpredictability)
if reportinfo.totalaccuracy is None:
reportinfo.totalaccuracy = ge.Test_Obj(net, data["data"], config["comparison-mode"]) * 100 /len(data["data"])
localscore = best*100 / len(data["data"])
out += ("<p>This gets " + str(round(localscore, 2)) + "% of the data items correct." +
"This is " + str(round(localscore * 100 / reportinfo.totalaccuracy, 2)) + "% of the max accuracy.</p>\n")
reportinfo.usedpredictability.append(xaxis)
reportinfo.usedpredictability.append(yaxis)
return out
def Graph(config):
netfile = config["net-file"]
net = gn.loadNets(netfile)[0][0]
usedata = config["use-data"]
xaxis = config["xaxis"]
yaxis = config["yaxis"]
zaxis = config["zaxis"]
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
if usedata:
datafile = config["data-file"]
colormode = config["color-mode"]
useclumping = config["clump"]
usepercents = config["usepercents"]
with open(datafile) as f:
data = json.load(f)
MLGraph.GraphNetData(net, data, xaxis, yaxis, zaxis, 0, useclumping,
usepercents, colormode, plt)
else:
MLGraph.GraphNet(net, xaxis, yaxis, plt)
pyplot.show()
def GenerateHighVarianceGraph(net, data, config, directory, fname):
global reportinfo
#region Measure Variance
best = 0
secondbest = 0
xaxis = ""
yaxis = ""
for inputa in net.inputs:
if inputa in reportinfo.usedvariance:
continue
outputs = []
x = -1
while x <= 1:
net.reset()
for name in net.inputs:
if name == inputa:
net.setNode(name, x)
else:
net.setNode(name, 0)
net.process()
outputs.append(net.getOutput()[net.classifier_output])
x += 0.1
dif = max(outputs) - min(outputs)
if dif > best:
yaxis = xaxis
xaxis = inputa
secondbest = best
best = dif
elif dif > secondbest:
yaxis = inputa
secondbest = dif
if yaxis == "":
return "<p>Not enough variables to create graph.</p>\n"
#endregion
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
if not config["add-data"]:
MLGraphing.GraphNet(net, xaxis, yaxis, plt)
else:
MLGraphing.GraphNetData(net, data, xaxis, yaxis, net.classifier_output, 0, config["clump"], config["percents"], config["color"], plt)
if config["customize"]:
pyplot.show()
fig.savefig(directory + '/' + fname, bbox_inches='tight')
pyplot.close(fig)
out = GenerateGraphHTML(net, config["header"], fname, xaxis, yaxis, config["add-data"], config["color"],
config["clump"], config["percents"], reportinfo.usedvariance)
reportinfo.usedvariance.append(xaxis)
reportinfo.usedvariance.append(yaxis)
return out
def creategraph(fig,
net,
inputdict,
outputdict,
title,
defvalue=0,
data=None,
percents=False,
colormode=0,
clump=False):
#we can assume that inputs = 2, and outputs > 1:
#region scan inputs and outputs
xaxis = ""
yaxis = ""
for name in inputdict:
if inputdict[name].state:
if xaxis == "":
xaxis = name
elif yaxis == "":
yaxis = name
else:
print("Error, too many inputs")
outputs = []
for name in outputdict:
if outputdict[name].state:
outputs.append(name)
if data != {} and data is not None:
grapher.GraphNetData(net,
data,
xaxis,
yaxis,
net.classifier_output,
defvalue,
clump,
percents,
colormodes[colormode],
fig,
title=title)
else:
grapher.GraphNetOutputs(net, xaxis, yaxis, outputs, fig, title=title)
def GenerateCustomNetDataGraph(net, data, config, directory, fname):
global reportinfo
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
MLGraphing.GraphNetData(net, data, config["x"], config["y"], net.classifier_output, 0, config["clump"], config["percents"], config["color"], plt)
if config["customize"]:
pyplot.show()
fig.savefig(directory + '/' + fname, bbox_inches='tight')
pyplot.close(fig)
out = GenerateGraphHTML(net, config["header"], fname, config["x"], config["y"], True, config["color"], config["clump"],
config["percents"], [])
return out
def GenerateCustomNetGraph(net, config, directory, fname):
global reportinfo
if config["x"] is None or config["y"] is None:
warnings.warn("Custom net graph missing configuration")
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
MLGraphing.GraphNet(net, config["x"], config["y"], plt)
if config["customize"]:
pyplot.show()
fig.savefig(directory + '/' + fname, bbox_inches = 'tight')
pyplot.close(fig)
out = GenerateGraphHTML(net, config["header"], fname, config["x"], config["y"], False, "", False, False, [])
return out
def GenerateNonLinearVarianceGraph(net, data, config, directory, fname):
#region measure variance
variance = []
nonlinear = []
inputs = []
for inputa in net.inputs:
if inputa in reportinfo.nonlinear:
continue
outputs = []
x = -1
while x <= 1:
net.reset()
for name in net.inputs:
if name == inputa:
net.setNode(name, x)
else:
net.setNode(name, 0)
net.process()
outputs.append(net.getOutput()[net.classifier_output])
x += 0.1
variance.append(max(outputs)-min(outputs))
inputs.append(inputa)
nonlinear.append((3 < outputs.index(max(outputs)) < 18 or 3 < outputs.index(min(outputs)) < 18))
#endregion
#region find nonlinear
count = 0
for item in nonlinear:
if item:
count += 1
target = 0
secondtarget = 0
xaxis = ""
yaxis = ""
bonustext = ""
if count == 0:
out = "<h3>" + config["header"] + "</h3>\n<p>None found"
if len(reportinfo.nonlinear) > 0:
out += "(excluding: " + ', '.join(reportinfo.nonlinear) +")"
out += "</p>\n"
return out
if count == 2:
for i in range(0, len(nonlinear)):
if nonlinear[i] and xaxis == "":
xaxis = inputs[i]
else:
yaxis = inputs[i]
if count > 2:
for i in range(0, len(nonlinear)):
if nonlinear[i]:
if variance[i] > target:
yaxis = xaxis
secondtarget = target
xaxis = inputs[i]
target = variance[i]
elif variance[i] > secondtarget:
yaxis = inputs[i]
secondtarget = variance[i]
bonustext = "<p>More than 2 non-linear inputs. We picked the ones with the most variance</p>\n"
if count == 1:
secondtarget = max(variance)
for i in range(0, len(nonlinear)):
if nonlinear[i]:
xaxis = inputs[i]
target = variance[i]
for i in range(0, len(nonlinear)):
if inputs[i] != xaxis and abs(variance[i] - target) < secondtarget:
secondtarget = abs(variance[i] - target)
yaxis = inputs[i]
bonustext= "<p>Only one non-linear input (" + xaxis +"). We picked a second input with similar variance.</p>\n"
#endregion
fig = pyplot.figure()
plt = fig.add_subplot(111, projection='3d')
if not config["add-data"]:
MLGraphing.GraphNet(net, xaxis, yaxis, plt)
else:
MLGraphing.GraphNetData(net, data, xaxis, yaxis, net.classifier_output, 0, config["clump"], config["percents"], config["color"], plt)
if config["customize"]:
pyplot.show()
fig.savefig(directory + '/' + fname, bbox_inches='tight')
pyplot.close(fig)
out = GenerateGraphHTML(net, config["header"], fname, xaxis, yaxis, config["add-data"], config["color"],
config["clump"], config["percents"], reportinfo.nonlinear)
reportinfo.nonlinear.append(xaxis)
reportinfo.nonlinear.append(yaxis)
return out + bonustext