PyTorch for Numpy users.

PyTorch version of Torch for Numpy users.
We assume you use the latest PyTorch and Numpy.

How to contribute?

git clone https://github.com/wkentaro/pytorch-for-numpy-users.git
cd pytorch-for-numpy-users
vim conversions.yaml
git commit -m "Update conversions.yaml"

./run_tests.py

Types

Numpy	PyTorch
np.ndarray	torch.Tensor
np.float32	torch.float32; torch.float
np.float64	torch.float64; torch.double
np.float16	torch.float16; torch.half
np.int8	torch.int8
np.uint8	torch.uint8
np.int16	torch.int16; torch.short
np.int32	torch.int32; torch.int
np.int64	torch.int64; torch.long

Ones and zeros

Numpy	PyTorch
np.empty((2, 3))	torch.empty(2, 3)
np.empty_like(x)	torch.empty_like(x)
np.eye	torch.eye
np.identity	torch.eye
np.ones	torch.ones
np.ones_like	torch.ones_like
np.zeros	torch.zeros
np.zeros_like	torch.zeros_like

From existing data

Numpy	PyTorch
np.array([[1, 2], [3, 4]])	torch.tensor([[1, 2], [3, 4]])
np.array([3.2, 4.3], dtype=np.float16) np.float16([3.2, 4.3])	torch.tensor([3.2, 4.3], dtype=torch.float16)
x.copy()	x.clone()
x.astype(np.float32)	x.type(torch.float32); x.float()
np.fromfile(file)	torch.tensor(torch.Storage(file))
np.frombuffer
np.fromfunction
np.fromiter
np.fromstring
np.load	torch.load
np.loadtxt
np.concatenate	torch.cat

Numerical ranges

Numpy	PyTorch
np.arange(10)	torch.arange(10)
np.arange(2, 3, 0.1)	torch.arange(2, 3, 0.1)
np.linspace	torch.linspace
np.logspace	torch.logspace

Linear algebra

Numpy	PyTorch
np.dot	torch.dot # 1D arrays only torch.mm # 2D arrays only torch.mv # matrix-vector (2D x 1D)
np.matmul	torch.matmul
np.tensordot	torch.tensordot
np.einsum	torch.einsum

Building matrices

Numpy	PyTorch
np.diag	torch.diag
np.tril	torch.tril
np.triu	torch.triu

Attributes

Numpy	PyTorch
x.shape	x.shape; x.size()
x.strides	x.stride()
x.ndim	x.dim()
x.data	x.data
x.size	x.nelement()
x.dtype	x.dtype

Indexing

Numpy	PyTorch
x[0]	x[0]
x[:, 0]	x[:, 0]
x[indices]	x[indices]
np.take(x, indices)	torch.take(x, torch.LongTensor(indices))
x[x != 0]	x[x != 0]

Shape manipulation

Numpy	PyTorch
x.reshape	x.reshape; x.view
x.resize()	x.resize_
	x.resize_as_
x = np.arange(6).reshape(3, 2, 1) x.transpose(2, 0, 1) # 012 -> 201	x = torch.arange(6).reshape(3, 2, 1) x.permute(2, 0, 1); x.transpose(1, 2).transpose(0, 1) # 012 -> 021 -> 201
x.flatten	x.view(-1)
x.squeeze()	x.squeeze()
x[:, None]; np.expand_dims(x, 1)	x[:, None]; x.unsqueeze(1)

Item selection and manipulation

Numpy	PyTorch
np.put
x.put	x.put_
x = np.array([1, 2, 3]) x.repeat(2) # [1, 1, 2, 2, 3, 3]	x = torch.tensor([1, 2, 3]) x.repeat_interleave(2) # [1, 1, 2, 2, 3, 3] x.repeat(2) # [1, 2, 3, 1, 2, 3] x.repeat(2).reshape(2, -1).transpose(1, 0).reshape(-1) # [1, 1, 2, 2, 3, 3]
np.tile(x, (3, 2))	x.repeat(3, 2)
x = np.array([[0, 1], [2, 3], [4, 5]]) idxs = np.array([0, 2]) np.choose(idxs, x) # [0, 5]	x = torch.tensor([[0, 1], [2, 3], [4, 5]]) idxs = torch.tensor([0, 2]) x[idxs, torch.arange(x.shape[1])] # [0, 5] torch.gather(x, 0, idxs[None, :])[0] # [0, 5]
np.sort	sorted, indices = torch.sort(x, [dim])
np.argsort	sorted, indices = torch.sort(x, [dim])
np.nonzero	torch.nonzero
np.where	torch.where
x[::-1]	torch.flip(x, [0])
np.unique(x)	torch.unique(x)

Calculation

Numpy	PyTorch
x.min	x.min
x.argmin	x.argmin
x.max	x.max
x.argmax	x.argmax
x.clip	x.clamp
x.round	x.round
np.floor(x)	torch.floor(x); x.floor()
np.ceil(x)	torch.ceil(x); x.ceil()
x.trace	x.trace
x.sum	x.sum
x.sum(axis=0)	x.sum(0)
x.cumsum	x.cumsum
x.mean	x.mean
x.std	x.std
x.prod	x.prod
x.cumprod	x.cumprod
x.all	x.all
x.any	x.any

Arithmetic and comparison operations

Numpy	PyTorch
np.less	x.lt
np.less_equal	x.le
np.greater	x.gt
np.greater_equal	x.ge
np.equal	x.eq
np.not_equal	x.ne

Random numbers

Numpy	PyTorch
np.random.seed	torch.manual_seed
np.random.permutation(5)	torch.randperm(5)

Numerical operations

Numpy	PyTorch
np.sign	torch.sign
np.sqrt	torch.sqrt