def evaluate(dAldKRLS, vX):
"""
Function evaluates the current krls algorithm. |br|
Args:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
vX (numpy array or integer): Evaluation argument
Returns:
vY (numpy array): x = alpha * dictionary, where 'alpha' and 'dictionary' are outcomes of the
training of the KRLS algporithm stored in the dAldKRLS dictionary
"""
# -------------------------------------------------------------------
# Check the input data
#
# Check if vX is a number? If so, make a 1x1 numpy array
if type(vX) is not np.ndarray:
vX = np.array([[vX]])
# Check if vX is a 1-dim np array? If so, make is a 2-dim numpy array
elif vX.ndim == 1:
vX = np.array([vX]).T
elif vX.ndim == 2:
pass
else:
strErr = (
'Input vector must be 1dim or 2dim numpy array or an integer')
raise ValueError(strErr)
# -------------------------------------------------------------------
# Get the needed data from the dictionary with data
strKernel = dAldKRLS['strKernel'] # Type of the kernel
iKernelPar = dAldKRLS['iKernelPar'] # Kernel parameter
mDict = dAldKRLS['mDict']
vAlpha = dAldKRLS['vAlpha']
if mDict.ndim == 2:
(iRowsDict, _) = mDict.shape # The number of rows in the dictionary
else:
(iRowsDict, ) = mDict.shape # The number of rows in the dictionary
if iRowsDict > 0:
vK = kernel.main(mDict, vX, strKernel, iKernelPar)
vY = np.dot(vK.T, vAlpha)
else:
vY = np.zeros((iRowsDict, 1))
return vY
def evaluate(dAldKRLS, vX):
"""
Function evaluates the current krls algorithm. |br|
Args:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
vX (numpy array or integer): Evaluation argument
Returns:
vY (numpy array): x = alpha * dictionary, where 'alpha' and 'dictionary' are outcomes of the
training of the KRLS algporithm stored in the dAldKRLS dictionary
"""
# -------------------------------------------------------------------
# Check the input data
#
# Check if vX is a number? If so, make a 1x1 numpy array
if type(vX) is not np.ndarray:
vX = np.array([[vX]])
# Check if vX is a 1-dim np array? If so, make is a 2-dim numpy array
elif vX.ndim == 1:
vX = np.array([vX]).T
elif vX.ndim == 2:
pass
else:
strErr = "Input vector must be 1dim or 2dim numpy array or an integer"
raise ValueError(strErr)
# -------------------------------------------------------------------
# Get the needed data from the dictionary with data
strKernel = dAldKRLS["strKernel"] # Type of the kernel
iKernelPar = dAldKRLS["iKernelPar"] # Kernel parameter
mDict = dAldKRLS["mDict"]
vAlpha = dAldKRLS["vAlpha"]
if mDict.ndim == 2:
(iRowsDict, _) = mDict.shape # The number of rows in the dictionary
else:
(iRowsDict,) = mDict.shape # The number of rows in the dictionary
if iRowsDict > 0:
vK = kernel.main(mDict, vX, strKernel, iKernelPar)
vY = np.dot(vK.T, vAlpha)
else:
vY = np.zeros((iRowsDict, 1))
return vY
def train(dAldKRLS, vX, vY):
"""
Function trains the current krls algorithm. |br|
Args:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
Must be initialized by the *init* function from this module.
vX (numpy array): X training data. It should be 2D or 1D numpy array or an integer.
vY (numpy array): Y trainig data. It should be 2D or 1D numpy array or an integer.
Returns:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
Fields in the output dictionary:
- a. **iALDth** (*int*): ALD threshold
- b. **iMaxDt** (*float*): Max size of the dictionary
- c. **strKernel** (*string*): Type of the kernel
- d. **iKernelPar** (*float*): Kernel parameter
- e. **bInit** (*int*): Initialization flag = 0. This flag is cleared with a first call to the 'train' function.
- f. **mKinv** (*numpy array*): Kernel matrix
- g. **vAlpha** (*numpy array*): Alpha vector
- h. **mP** (*numpy array*): 'P' matrix
- i. **mDict** (*numpy array*): Dictionary matrix
"""
# -------------------------------------------------------------------
# Check the input data
#
# Check if vX is a number? If so, make it a 1x1 numpy array
if type(vX) is not np.ndarray:
vX = np.array([[vX]])
# Check if vX is a 1-dim np array? If so, make it a 2-dim numpy array
elif vX.ndim == 1:
vX = np.array([vX])
elif vX.ndim == 2:
pass
else:
strErr = ('vX must be 1dim or 2dim numpy array or an integer')
raise ValueError(strErr)
# Check if vY is a number? If so, make it a 1x1 numpy array
if type(vY) is not np.ndarray:
vY = np.array([[vY]])
# Check if vY is a 1-dim np array? If so, make it a 2-dim numpy array
elif vY.ndim == 1:
vY = np.array([vY])
elif vY.ndim == 2:
pass
else:
strErr = ('vY must be 1dim or 2dim numpy array or an integer')
raise ValueError(strErr)
# -------------------------------------------------------------------
bInit = dAldKRLS['bInit'] # Get the initialization flag
if (bInit == 0):
# Initialize training:
# Get the needed data
strKernel = dAldKRLS['strKernel'] # Type of the kernel
iKernelPar = dAldKRLS['iKernelPar'] # Kernel parameter
vK = kernel.main(vX, vX.T, strKernel, iKernelPar)
vKt = vK[np.arange(0, vK.size - 1)]
iKtt = float(vK[vK.size - 1])
# Store all the parameters in the dictionary
dAldKRLS['mKinv'] = 1 / iKtt
dAldKRLS['vAlpha'] = vY / iKtt
dAldKRLS['mP'] = np.array([[1]])
dAldKRLS['mDict'] = vX
dAldKRLS['bInit'] = 1 # Set 'Initialization done' flag
else:
# Training:
# Get the needed data from the dictionary with data
strKernel = dAldKRLS['strKernel'] # Type of the kernel
iKernelPar = dAldKRLS['iKernelPar'] # Kernel parameter
mDict = dAldKRLS['mDict']
vAlpha = dAldKRLS['vAlpha']
mKinv = dAldKRLS['mKinv']
iALDth = dAldKRLS['iALDth'] # ALD threshold
iMaxDict = dAldKRLS['iMaxDt'] # Maximum size of the dictionary
mP = dAldKRLS['mP']
# Computations start here
vK = kernel.main(np.vstack((mDict, vX)), vX.T, strKernel, iKernelPar)
vKt = vK[np.arange(0, vK.size - 1)]
iKtt = float(vK[vK.size - 1])
vAT = np.dot(mKinv, vKt) # AT vector
iDelta = iKtt - float(np.dot(vKt.T, vAT)) # Delta value (integer)
if mDict.ndim == 2:
(iRowsDict,
_) = mDict.shape # The number of rows in the dictionary
else:
(iRowsDict, ) = mDict.shape # The number of rows in the dictionary
if ((iDelta > iALDth)
and (iRowsDict < iMaxDict)): # expand, if delta is higher than
# ALD threshold, and there is a
# place size of the dictionary
mDict = np.vstack((mDict, vX))
mKinv = (1 / iDelta) * np.vstack(
(np.hstack(((iDelta * mKinv + np.dot(vAT, vAT.T)), -vAT)),
np.hstack((-vAT.T, np.array([[1]])))))
(iRowsP, _) = mP.shape
vZ = np.zeros((iRowsP, 1))
mP = np.vstack((np.hstack(
(mP, vZ)), np.hstack((vZ.T, np.array([[1]])))))
iOde = 1 / iDelta * (vY - np.dot(vKt.T, vAlpha))
vAlpha = np.vstack(((vAlpha - np.dot(vAT, iOde)), iOde))
else: # only update alpha
vQ = np.dot(mP, vAT) / (1 + np.dot(np.dot(vAT.T, mP), vAT))
mP = mP - np.dot(vQ, (np.dot(vAT.T, mP)))
vAlpha = vAlpha + mKinv.dot(vQ).dot(vY - (vKt.T).dot(vAlpha))
# Store the data in the dictionary
dAldKRLS['mDict'] = mDict
dAldKRLS['vAlpha'] = vAlpha
dAldKRLS['mKinv'] = mKinv
dAldKRLS['mP'] = mP
return dAldKRLS
def train(dAldKRLS, vX, vY):
"""
Function trains the current krls algorithm. |br|
Args:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
Must be initialized by the *init* function from this module.
vX (numpy array): X training data. It should be 2D or 1D numpy array or an integer.
vY (numpy array): Y trainig data. It should be 2D or 1D numpy array or an integer.
Returns:
dAldKRLS (dictionary): Python dictionary which contains all the data of the current KRLS algorithm.
Fields in the output dictionary:
- a. **iALDth** (*int*): ALD threshold
- b. **iMaxDt** (*float*): Max size of the dictionary
- c. **strKernel** (*string*): Type of the kernel
- d. **iKernelPar** (*float*): Kernel parameter
- e. **bInit** (*int*): Initialization flag = 0. This flag is cleared with a first call to the 'train' function.
- f. **mKinv** (*numpy array*): Kernel matrix
- g. **vAlpha** (*numpy array*): Alpha vector
- h. **mP** (*numpy array*): 'P' matrix
- i. **mDict** (*numpy array*): Dictionary matrix
"""
# -------------------------------------------------------------------
# Check the input data
#
# Check if vX is a number? If so, make it a 1x1 numpy array
if type(vX) is not np.ndarray:
vX = np.array([[vX]])
# Check if vX is a 1-dim np array? If so, make it a 2-dim numpy array
elif vX.ndim == 1:
vX = np.array([vX])
elif vX.ndim == 2:
pass
else:
strErr = "vX must be 1dim or 2dim numpy array or an integer"
raise ValueError(strErr)
# Check if vY is a number? If so, make it a 1x1 numpy array
if type(vY) is not np.ndarray:
vY = np.array([[vY]])
# Check if vY is a 1-dim np array? If so, make it a 2-dim numpy array
elif vY.ndim == 1:
vY = np.array([vY])
elif vY.ndim == 2:
pass
else:
strErr = "vY must be 1dim or 2dim numpy array or an integer"
raise ValueError(strErr)
# -------------------------------------------------------------------
bInit = dAldKRLS["bInit"] # Get the initialization flag
if bInit == 0:
# Initialize training:
# Get the needed data
strKernel = dAldKRLS["strKernel"] # Type of the kernel
iKernelPar = dAldKRLS["iKernelPar"] # Kernel parameter
vK = kernel.main(vX, vX.T, strKernel, iKernelPar)
vKt = vK[np.arange(0, vK.size - 1)]
iKtt = float(vK[vK.size - 1])
# Store all the parameters in the dictionary
dAldKRLS["mKinv"] = 1 / iKtt
dAldKRLS["vAlpha"] = vY / iKtt
dAldKRLS["mP"] = np.array([[1]])
dAldKRLS["mDict"] = vX
dAldKRLS["bInit"] = 1 # Set 'Initialization done' flag
else:
# Training:
# Get the needed data from the dictionary with data
strKernel = dAldKRLS["strKernel"] # Type of the kernel
iKernelPar = dAldKRLS["iKernelPar"] # Kernel parameter
mDict = dAldKRLS["mDict"]
vAlpha = dAldKRLS["vAlpha"]
mKinv = dAldKRLS["mKinv"]
iALDth = dAldKRLS["iALDth"] # ALD threshold
iMaxDict = dAldKRLS["iMaxDt"] # Maximum size of the dictionary
mP = dAldKRLS["mP"]
# Computations start here
vK = kernel.main(np.vstack((mDict, vX)), vX.T, strKernel, iKernelPar)
vKt = vK[np.arange(0, vK.size - 1)]
iKtt = float(vK[vK.size - 1])
vAT = np.dot(mKinv, vKt) # AT vector
iDelta = iKtt - float(np.dot(vKt.T, vAT)) # Delta value (integer)
if mDict.ndim == 2:
(iRowsDict, _) = mDict.shape # The number of rows in the dictionary
else:
(iRowsDict,) = mDict.shape # The number of rows in the dictionary
if (iDelta > iALDth) and (iRowsDict < iMaxDict): # expand, if delta is higher than
# ALD threshold, and there is a
# place size of the dictionary
mDict = np.vstack((mDict, vX))
mKinv = (1 / iDelta) * np.vstack(
(np.hstack(((iDelta * mKinv + np.dot(vAT, vAT.T)), -vAT)), np.hstack((-vAT.T, np.array([[1]]))))
)
(iRowsP, _) = mP.shape
vZ = np.zeros((iRowsP, 1))
mP = np.vstack((np.hstack((mP, vZ)), np.hstack((vZ.T, np.array([[1]])))))
iOde = 1 / iDelta * (vY - np.dot(vKt.T, vAlpha))
vAlpha = np.vstack(((vAlpha - np.dot(vAT, iOde)), iOde))
else: # only update alpha
vQ = np.dot(mP, vAT) / (1 + np.dot(np.dot(vAT.T, mP), vAT))
mP = mP - np.dot(vQ, (np.dot(vAT.T, mP)))
vAlpha = vAlpha + mKinv.dot(vQ).dot(vY - (vKt.T).dot(vAlpha))
# Store the data in the dictionary
dAldKRLS["mDict"] = mDict
dAldKRLS["vAlpha"] = vAlpha
dAldKRLS["mKinv"] = mKinv
dAldKRLS["mP"] = mP
return dAldKRLS
"""
#!/usr/bin/python3
"""
"""
Have a good time reading the source. You are an amazing person.
"""
from kernel.main import *
from kernel.other import *
from kernel.banner import pbanner
import _thread
def main():
clear_screen()
print(pbanner())
Console()
main()