How to use add_kernel method in autotest

Best Python code snippet using autotest_python

models.py

Source:models.py

1import gpytorch2from gpytorch import lazify3from gpytorch.models import ExactGP, AbstractVariationalGP4from gpytorch.variational import CholeskyVariationalDistribution, VariationalStrategy5import torch6from gp_models.kernels import GeneralizedProjectionKernel7from gp_models import CustomAdditiveKernel8class ExactGPModel(ExactGP):9    """Basic exact GP model with const mean and a provided kernel"""10    def __init__(self, train_x, train_y, likelihood, kernel):11        super(ExactGPModel, self).__init__(train_x, train_y, likelihood)12        self.mean_module = gpytorch.means.ConstantMean()13        self.covar_module = kernel14    def forward(self, x):15        mean_x = self.mean_module(x)16        covar_x = self.covar_module(x)17        return gpytorch.distributions.MultivariateNormal(mean_x, covar_x)18class AdditiveExactGPModel(ExactGPModel):19    def __init__(self, train_x, train_y, likelihood, kernel):20        if isinstance(kernel, gpytorch.kernels.ScaleKernel):21            if not isinstance(kernel.base_kernel, CustomAdditiveKernel):22                raise ValueError("Not an additive kernel.")23        else:24            if not isinstance(kernel, CustomAdditiveKernel):25                raise ValueError("Not an additive kernel.")26        super(AdditiveExactGPModel, self).__init__(train_x, train_y, likelihood, kernel)27    def additive_pred(self, x, group=None):  # Pretty sure this should count as a prediction strategy28        # TODO: implement somehow for RP models as well.29        if isinstance(self.covar_module, gpytorch.kernels.ScaleKernel):30            scale = self.covar_module.outputscale31            add_kernel = self.covar_module.base_kernel32        elif isinstance(self.covar_module,  CustomAdditiveKernel):33            scale = torch.tensor(1.0, dtype=torch.float)34            add_kernel = self.covar_module35        else:36            raise NotImplementedError("Only implemented for Additive kernels and Scale kernel wrappings only")37        train_prior_dist = self.forward(self.train_inputs[0])38        lik_covar_train_train = self.likelihood(train_prior_dist, self.train_inputs[0]).lazy_covariance_matrix.detach()39        # TODO: cache?40        K_inv_y = lik_covar_train_train.inv_matmul(self.train_targets)41        def get_pred(k):42            # is there a better way to handle the scalings?43            test_train_covar = scale * gpytorch.lazy.delazify(k(x, self.train_inputs[0]))44            test_test_covar = scale * gpytorch.lazy.delazify(k(x, x))45            train_test_covar = test_train_covar.transpose(-1, -2)46            pred_mean = test_train_covar.matmul(K_inv_y)47            covar_correction_rhs = lik_covar_train_train.inv_matmul(train_test_covar)48            pred_covar = lazify(torch.addmm(1, test_test_covar, -1, test_train_covar, covar_correction_rhs))49            return gpytorch.distributions.MultivariateNormal(pred_mean, pred_covar, validate_args=False)50        if group is None:51            outputs = []52            for k in add_kernel.kernel.kernels:53                outputs.append(get_pred(k))54            return outputs55        else:56            return get_pred(add_kernel.kernel.kernels[group])57    def get_groups(self):58        if isinstance(self.covar_module, gpytorch.kernels.ScaleKernel):59            add_kernel = self.covar_module.base_kernel60        else:61            add_kernel = self.covar_module62        return add_kernel.groups63class ProjectedAdditiveExactGPModel(ExactGPModel):64    def __init__(self, train_x, train_y, likelihood, kernel):65        if isinstance(kernel, gpytorch.kernels.ScaleKernel):66            if not isinstance(kernel.base_kernel, GeneralizedProjectionKernel):67                raise ValueError("Not an projected additive kernel.")68        else:69            if not isinstance(kernel, GeneralizedProjectionKernel):70                raise ValueError("Not an projected additive kernel.")71        super(ProjectedAdditiveExactGPModel, self).__init__(train_x, train_y, likelihood, kernel)72    def get_corresponding_additive_model(self, return_proj=True):73        return convert_rp_model_to_additive_model(self, return_proj=return_proj)74class SVGPRegressionModel(AbstractVariationalGP):75    """SVGP with provided kernel, Cholesky variational prior, and provided inducing points."""76    def __init__(self, inducing_points, kernel, likelihood):77        variational_distribution = CholeskyVariationalDistribution(78            inducing_points.size(0))79        variational_strategy = VariationalStrategy(self,80                                                   inducing_points,81                                                   variational_distribution,82                                                   learn_inducing_locations=True)83        super(SVGPRegressionModel, self).__init__(variational_strategy)84        self.mean_module = gpytorch.means.ConstantMean()85        self.covar_module = kernel86        self.likelihood = likelihood87    def forward(self,x):88        mean_x = self.mean_module(x)89        covar_x = self.covar_module(x)90        latent_pred = gpytorch.distributions.MultivariateNormal(mean_x, covar_x)91        return latent_pred92def convert_rp_model_to_additive_model(model, return_proj=True):93    if isinstance(model.covar_module, gpytorch.kernels.ScaleKernel):94        rp_kernel = model.covar_module.base_kernel95    else:96        rp_kernel = model.covar_module97    add_kernel = rp_kernel.to_additive_kernel()98    proj = rp_kernel.projection_module99    if isinstance(model.covar_module, gpytorch.kernels.ScaleKernel):100        add_kernel = gpytorch.kernels.ScaleKernel(add_kernel)101        add_kernel.initialize(outputscale=model.covar_module.outputscale)102    Z = rp_kernel.projection_module(model.train_inputs[0])103    res = AdditiveExactGPModel(Z, model.train_targets, model.likelihood, add_kernel)104    res.mean_module = model.mean_module  # either zero or constant, so doesn't matter if the input is projected.105    if return_proj:106        res = (res, proj)...

4. cuda_kernels.py

Source:4. cuda_kernels.py

...19# each grid has blocks on threads20# Each thread is identified by a threadidx (index for threads)21# and blockidx(index for the block)22@cuda.jit23def add_kernel(x,y,out):24    tx=cuda.threadIdx.x25    ty=cuda.blockIdx.y26    block_size=cuda.blockDim.x # num threads per block27    grid_size=cuda.gridDim.x # num blocks in the grid28    start=tx+ty*block_size29    stride=block_size*grid_size30    for i in range(start,x.shape[0],stride):31        out[i]=x[i]+y[i]32n=100000033x=np.arange(n).astype(np.float32)34y=2*x35out=np.empty_like(x)36threads_per_block=12837blocks_per_grid=3038add_kernel[blocks_per_grid,threads_per_block](x,y,out)39out[:10]40# Using Numba helper functions41@cuda.jit42def add_kernel(x,y,out):43    start=cuda.grid(1) # 1-d thread grid44    stride=cuda.gridsize(1) # 1-d thread grid45    for i in range(start,x.shape[0],stride):46        out[i]=x[i]+y[i]47x_device=cuda.to_device(x)48y_device=cuda.to_device(y)49out_device=cuda.device_array_like(x)50start=perf_counter()51add_kernel[threads_per_block,blocks_per_grid](x,y,out)52print("Time: "+repr(perf_counter()-start)+"s")53start=perf_counter()54add_kernel[threads_per_block,blocks_per_grid](x_device,y_device,out_device)55print("Time: "+repr(perf_counter()-start)+"s")56out_device.copy_to_host()...

elwise_gfuncs.py

Source:elwise_gfuncs.py

...18    __all__.append(root)19    for r in kernel_roots:20        for t in types:21            name = r + '_' + t22            f.add_kernel(elwise_kernels.__dict__[name])23types = ['int32', 'int64', 'uint32', 'uint64', 'float32', 'float64']24add_basic_gfunc('add', types)25add_basic_gfunc('subtract', types)26add_basic_gfunc('multiply', types)27add_basic_gfunc('divide', types)28#add_basic_gfunc('maximum', types, ['maximum2', 'maximum3'])29#add_basic_gfunc('minimum', types, ['minimum2', 'minimum3'])30add_basic_gfunc('maximum', types, ['maximum2'])31add_basic_gfunc('minimum', types, ['minimum2'])32add_basic_gfunc('square', types)33add_basic_gfunc('abs', types)34add_basic_gfunc('floor', types)35add_basic_gfunc('ceil', types)36if sys.platform == 'win32':37    fmod = gfunc.elwise('fmod')38    fmod.add_kernel(elwise_kernels.fmod)39    pow = gfunc.elwise('pow')40    pow.add_kernel(elwise_kernels.pow)41    sqrt = gfunc.elwise('sqrt')42    sqrt.add_kernel(elwise_kernels.sqrt)43    exp = gfunc.elwise('exp')44    exp.add_kernel(elwise_kernels.exp)45    log = gfunc.elwise('log')46    log.add_kernel(elwise_kernels.log)47    log10 = gfunc.elwise('log10')48    log10.add_kernel(elwise_kernels.log10)49    sin = gfunc.elwise('sin')50    sin.add_kernel(elwise_kernels.sin)51    cos = gfunc.elwise('cos')52    cos.add_kernel(elwise_kernels.cos)53    tan = gfunc.elwise('tan')54    tan.add_kernel(elwise_kernels.tan)55    arcsin = gfunc.elwise('arcsin')56    arcsin.add_kernel(elwise_kernels.arcsin)57    arccos = gfunc.elwise('arccos')58    arccos.add_kernel(elwise_kernels.arccos)59    arctan = gfunc.elwise('arctan')60    arctan.add_kernel(elwise_kernels.arctan)61    arctan2 = gfunc.elwise('arctan2')62    arctan2.add_kernel(elwise_kernels.arctan2)63    sinh = gfunc.elwise('sinh')64    sinh.add_kernel(elwise_kernels.sinh)65    cosh = gfunc.elwise('cosh')66    cosh.add_kernel(elwise_kernels.cosh)67    tanh = gfunc.elwise('tanh')68    tanh.add_kernel(elwise_kernels.tanh)69    ldexp = gfunc.elwise('ldexp')70    ldexp.add_kernel(elwise_kernels.ldexp)71    isnan = gfunc.elwise('isnan')72    isnan.add_kernel(elwise_kernels.isnan)73    isfinite = gfunc.elwise('isfinite')74    isfinite.add_kernel(elwise_kernels.isfinite)75    nextafter = gfunc.elwise('nextafter')...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.