tessl/pypi-keras-nightly
Multi-backend deep learning framework providing a unified API for building and training neural networks across JAX, TensorFlow, PyTorch, and OpenVINO backends
—
Pending
operations.mddocs/
Mathematical Operations
Backend-agnostic mathematical operations providing NumPy-compatible APIs, neural network specific functions, and core tensor operations that work across JAX, TensorFlow, PyTorch, and OpenVINO backends.
Capabilities
Core Operations
Essential tensor operations and control flow functions.
def cast(x, dtype):
"""Cast tensor to specified data type."""
def convert_to_tensor(x, dtype=None, sparse=None):
"""Convert input to backend tensor."""
def convert_to_numpy(x):
"""Convert tensor to NumPy array."""
def shape(x):
"""Get tensor shape as tuple."""
def dtype(x):
"""Get tensor data type."""
def stop_gradient(x):
"""Stop gradient computation through tensor."""
def cond(pred, true_fn, false_fn):
"""Conditional execution based on predicate."""
def scan(fn, init, xs, length=None, reverse=False, unroll=1):
"""Apply function across sequence elements."""
def map(fn, xs):
"""Map function over tensor elements."""NumPy-Compatible Operations
Mathematical functions matching NumPy API for cross-backend compatibility.
# Array creation
def zeros(shape, dtype='float32'): ...
def ones(shape, dtype='float32'): ...
def eye(N, M=None, k=0, dtype='float32'): ...
def arange(start, stop=None, step=1, dtype=None): ...
# Mathematical operations
def add(x1, x2): ...
def subtract(x1, x2): ...
def multiply(x1, x2): ...
def divide(x1, x2): ...
def power(x1, x2): ...
def sqrt(x): ...
def exp(x): ...
def log(x): ...
def sin(x): ...
def cos(x): ...
def tanh(x): ...
# Reduction operations
def sum(x, axis=None, keepdims=False): ...
def mean(x, axis=None, keepdims=False): ...
def max(x, axis=None, keepdims=False): ...
def min(x, axis=None, keepdims=False): ...
# Linear algebra
def matmul(x1, x2): ...
def dot(x1, x2): ...
def transpose(x, axes=None): ...
def reshape(x, shape): ...
# Indexing and manipulation
def concatenate(arrays, axis=0): ...
def stack(arrays, axis=0): ...
def split(x, indices_or_sections, axis=0): ...
def squeeze(x, axis=None): ...
def expand_dims(x, axis): ...Neural Network Operations
Specialized operations for neural networks and deep learning.
# Activation functions
def relu(x): ...
def sigmoid(x): ...
def softmax(x, axis=-1): ...
def gelu(x, approximate=False): ...
def silu(x): ... # Also available as swish
# Convolution operations
def conv(inputs, kernel, strides=1, padding='valid', data_format=None,
dilation_rate=1, groups=1): ...
def conv_transpose(inputs, kernel, strides=1, padding='valid'): ...
def depthwise_conv(inputs, kernel, strides=1, padding='valid'): ...
# Pooling operations
def max_pool(inputs, pool_size, strides=None, padding='valid'): ...
def average_pool(inputs, pool_size, strides=None, padding='valid'): ...
# Normalization
def batch_normalization(x, mean, variance, offset, scale, epsilon=1e-3): ...
def layer_normalization(x, axis=-1, epsilon=1e-3): ...
# Loss functions
def binary_crossentropy(target, output, from_logits=False): ...
def categorical_crossentropy(target, output, from_logits=False, axis=-1): ...
def sparse_categorical_crossentropy(target, output, from_logits=False, axis=-1): ...
# Utility functions
def one_hot(x, num_classes, axis=-1, dtype='float32'): ...
def top_k(x, k, sorted=True): ...
def in_top_k(targets, predictions, k): ...Linear Algebra Operations
Advanced linear algebra operations for machine learning algorithms.
def cholesky(x): ...
def det(x): ...
def eig(x): ...
def eigh(x): ...
def inv(x): ...
def lstsq(a, b, rcond=None): ...
def norm(x, ord=None, axis=None, keepdims=False): ...
def qr(x, mode='reduced'): ...
def solve(a, b): ...
def svd(x, full_matrices=True): ...FFT Operations
Fast Fourier Transform operations for signal processing.
def fft(x): ...
def fft2(x): ...
def rfft(x): ...
def irfft(x): ...
def stft(x, frame_length, frame_step, fft_length=None): ...
def istft(stfts, frame_length, frame_step, fft_length=None): ...Usage Examples
Basic Tensor Operations
import keras
from keras import ops
# Create tensors
x = ops.ones((3, 4))
y = ops.zeros((3, 4))
# Mathematical operations
z = ops.add(x, y)
w = ops.matmul(x, ops.transpose(y))
# Reductions
mean_val = ops.mean(x)
sum_val = ops.sum(x, axis=1)Neural Network Forward Pass
import keras
from keras import ops
def forward_pass(x, weights, bias):
# Linear transformation
x = ops.matmul(x, weights) + bias
# Activation
x = ops.relu(x)
# Normalization
mean = ops.mean(x, axis=0, keepdims=True)
variance = ops.var(x, axis=0, keepdims=True)
x = (x - mean) / ops.sqrt(variance + 1e-7)
return xInstall with Tessl CLI
npx tessl i tessl/pypi-keras-nightly