def train(self_,
X,
y,
solution=None,
iterations=20,
batchSize=64,
kerasArgs={},
**kwArgs):
shaper = solution.shaper() if solution else ma.DataShaper(X, **kwArgs)
sampleSize = int(sqrt(shaper.feature_count()))
model = kwArgs.get(
"model", KerasSolver.__prepare_model(shaper, y, solution,
**kwArgs))
y = shaper.map_labels(y)
y = keras.utils.to_categorical(y, num_classes=shaper.class_count())
X = shaper.conform(X, addOnes=False)
X = X.reshape(X.shape[0], sampleSize, sampleSize, 1)
X = X.astype('float32')
with NoWarnings():
model.fit(X,
y,
batch_size=batchSize,
epochs=iterations,
**kerasArgs)
return ma.Solution(model=model, shaper=shaper)
def train(self_, X, y, solution=None, Lambda=0, iterations=50, **kwArgs):
shaper = solution.shaper() if solution else ma.DataShaper(X, **kwArgs)
debug = kwArgs.get("debug", False)
topology, theta = NeuralNetworkSolver.__initial_theta(
shaper, y, solution, **kwArgs)
y = shaper.map_labels(y)
X = shaper.conform(X, addOnes=False)
theta = optimize.fmin_cg(compute_cost,
theta,
fprime=compute_grad,
args=(X, y, topology, shaper.class_count(),
Lambda, debug),
maxiter=iterations,
disp=False)
return ma.Solution(model=(topology, theta), shaper=shaper)
def train(self_, X, y, solution=None, Lambda=0, iterations=50, **kwArgs):
shaper = solution.shaper() if solution else ma.DataShaper(X, **kwArgs)
theta = LogisticRegressionSolver.__initial_theta(shaper,
y,
solution=solution,
**kwArgs)
y = shaper.map_labels(y)
X = shaper.conform(X)
for c in range(shaper.class_count()):
theta[c] = optimize.fmin_cg(compute_cost,
theta[c],
fprime=compute_grad,
args=(X, (y == c), Lambda),
maxiter=iterations,
disp=False)
return ma.Solution(model=theta, shaper=shaper)
def train(args):
solver = preferredEngine()
Lambda = list(map(float, args.Lambda.split(",")))
rawData = dio.load(args.data_set)
solution = None
try:
solution = dio.load(args.solution)
except:
pass
X_orig = rawData["X"]
y_orig = rawData["y"]
args.engine = solver.__class__.__name__
optimizationResults = ma.find_solution(solver,
X_orig,
y_orig,
solutionInit=solution,
showFailureRateTrain=True,
optimizationParams={
"nnTopology": args.topology,
"Lambda": Lambda
},
files=[args.data_set],
log={"log_file_name": "mlak.log"},
**vars(args))
solution = optimizationResults.solution
dio.save(args.solution, solution)
if args.debug: # pragma: no cover
print("solution = {}".format(solution))
if args.verbose and solution.shaper().is_classifier(): # pragma: no cover
for i in range(len(y_orig)):
if not predict(solver, solution, X_orig, y_orig, i):
return
def train(self_, X, y, solution=None, Lambda=0, iterations=50, **kwArgs):
shaper = solution.shaper() if solution else ma.DataShaper(X, **kwArgs)
theta = LinearRegressionSolver.__initial_theta(shaper, solution,
**kwArgs)
X = shaper.conform(X)
theta = oa.gradient_descent_fminCG(oa.Algorithm(
compute_cost, compute_grad),
X,
y,
theta,
iterations,
Lambda,
disp=False)
return ma.Solution(model=theta, shaper=shaper)
def analyze(args): # pragma: no cover
solver = preferredEngine()
rawData = dio.load(args.data_set)
X_orig = rawData["X"]
y_orig = rawData["y"]
args.engine = solver.__class__.__name__
analyzerResults = ma.analyze(solver,
X_orig,
y_orig,
optimizationParams={
"nnTopology": args.topology,
"Lambda": args.Lambda,
"functions": None
},
files=[args.data_set],
**vars(args))
fig = plt.figure(1)
plt.rc('grid', linestyle=":", color='gray')
plt.subplot(211)
plt.title("data: {}\nlambda: {}\n\n".format(args.data_set, args.Lambda),
loc="left")
plt.title("Sample count test\n")
plt.xlabel("Sample count")
plt.ylabel("Error rate")
plt.plot(analyzerResults.sampleCountAnalyzis.sampleCount,
analyzerResults.sampleCountAnalyzis.errorTrain,
'b-',
label="train")
plt.plot(analyzerResults.sampleCountAnalyzis.sampleCount,
analyzerResults.sampleCountAnalyzis.errorCV,
'g-',
label="CV")
plt.grid()
plt.legend()
plt.subplot(212)
plt.title("Iteration count test")
plt.xlabel("Iteration count")
plt.ylabel("Error rate")
plt.plot(analyzerResults.iterationCountAnalyzis.iterationCount,
analyzerResults.iterationCountAnalyzis.errorTrain,
'b-',
label="train")
plt.plot(analyzerResults.iterationCountAnalyzis.iterationCount,
analyzerResults.iterationCountAnalyzis.errorCV,
'g-',
label="CV")
plt.grid()
plt.legend()
def on_resize(event):
plt.tight_layout()
plt.subplots_adjust(right=0.95)
cid = fig.canvas.mpl_connect("resize_event", on_resize)
plt.show()
def examine(self_, solution, X):
yp = self_.__examine(solution, X)
return ma.label_confidence(solution.shaper(), yp)
def examine(self_, solution, X):
X = solution.shaper().conform(X)
yp = predict_one_vs_all(X, solution.model())
return ma.label_confidence(solution.shaper(), yp)