Data Core

Main Numpy and Times Series functions used throughout the library.

from tsai.data.external import get_UCR_data

dsid = 'OliveOil'
X_train, y_train, X_valid, y_valid = get_UCR_data(dsid, on_disk=True, force_download=True)
X_on_disk, y_on_disk, splits = get_UCR_data(dsid, on_disk=True, return_split=False, force_download=True)
X_in_memory, y_in_memory, splits = get_UCR_data(dsid, on_disk=False, return_split=False, force_download=True)
y_tensor = cat2int(y_on_disk)
y_array = y_tensor.numpy()

---

source

ToNumpyTensor

 ToNumpyTensor (enc=None, dec=None, split_idx=None, order=None)

Transforms an object into NumpyTensor

---

source

NumpyTensor

 NumpyTensor (o, dtype=None, device=None, copy=None, requires_grad=False,
              **kwargs)

Returns a tensor with subclass NumpyTensor that has a show method

---

source

TSTensor

 TSTensor (o, dtype=None, device=None, copy=None, requires_grad=False,
           **kwargs)

Returns a tensor with subclass TSTensor that has a show method

---

source

show_tuple

 show_tuple (tup, nrows:int=1, ncols:int=1,
             sharex:"bool|Literal['none','all','row','col']"=False,
             sharey:"bool|Literal['none','all','row','col']"=False,
             squeeze:bool=True, width_ratios:Sequence[float]|None=None,
             height_ratios:Sequence[float]|None=None,
             subplot_kw:dict[str,Any]|None=None,
             gridspec_kw:dict[str,Any]|None=None)

Display a timeseries plot from a decoded tuple

---

source

ToTSTensor

 ToTSTensor (enc=None, dec=None, split_idx=None, order=None)

Transforms an object into TSTensor

a = np.random.randn(2, 3, 4).astype(np.float16)
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

a = np.random.randn(2, 3, 4).astype(np.float32)
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

a = np.random.randint(10, size=10).astype(np.int64)
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

a = np.random.randint(10, size=10).astype(np.int32)
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

a = torch.rand(2, 3, 4).float()
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

a = torch.randint(3, (10,))
assert np.shares_memory(a, NumpyTensor(a))
assert np.shares_memory(a, TSTensor(a))

t = TSTensor(torch.randn(2, 3, 4))
p = torch.tensor(3., requires_grad=True)
test = torch.add(t, p)
test_eq(test.requires_grad, True)
test_eq(type(t.data), torch.Tensor)
test_eq(type(t), TSTensor)

l = L([0,1,2,3], [4,5,6,7], [8, 9, 10, 11])
TSTensor(l), TSTensor(l).data

(TSTensor(vars:3, len:4, device=cpu, dtype=torch.int64),
 tensor([[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11]]))

t = TSTensor(X_train)
for i in range(4):
    print(t, t.ndim, torch.is_tensor(t))
    if i < 3: t = t[0]

TSTensor(samples:30, vars:1, len:570, device=cpu, dtype=torch.float32) 3 True
TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32) 2 True
TSTensor(len:570, device=cpu, dtype=torch.float32) 1 True
TSTensor([-0.6113752722740173], device=cpu, dtype=torch.float32) 0 True

TSTensor(X_on_disk)

TSTensor(samples:60, vars:1, len:570, device=cpu, dtype=torch.float32)

ToTSTensor()(X_on_disk)

TSTensor(samples:60, vars:1, len:570, device=cpu, dtype=torch.float32)

TSTensor(X_train).show();

TSTensor(X_train).show(title='1');

show_tuple((TSTensor(X_train), ['1', '2']))

show_tuple((TSTensor(np.arange(10).reshape(2,5)), 1))

show_tuple((TSTensor(np.arange(10).reshape(2,5)), '1'))

show_tuple((TSTensor(np.arange(10).reshape(2,5)), [1,2]))

show_tuple((TSTensor(np.arange(10).reshape(2,5)), ['1', '2']))

---

source

TSMaskTensor

 TSMaskTensor (o, dtype=None, device=None, copy=None, requires_grad=False,
               **kwargs)

Returns a tensor with subclass NumpyTensor that has a show method

---

source

TSLabelTensor

 TSLabelTensor (o, dtype=None, device=None, copy=None,
                requires_grad=False, **kwargs)

Returns a tensor with subclass NumpyTensor that has a show method

t = TSLabelTensor(torch.randint(0,10,(1, 2, 3)))
t, t[0], t[0][0], t[0][0][0]

(TSLabelTensor(shape:(1, 2, 3), device=cpu, dtype=torch.int64),
 TSLabelTensor(shape:(2, 3), device=cpu, dtype=torch.int64),
 TSLabelTensor(shape:(3,), device=cpu, dtype=torch.int64),
 3)

t = TSMaskTensor(torch.randint(0,10,(1, 2, 3)))
t, t[0], t[0][0], t[0][0][0]

(TSMaskTensor(shape:(1, 2, 3), device=cpu, dtype=torch.int64),
 TSMaskTensor(shape:(2, 3), device=cpu, dtype=torch.int64),
 TSMaskTensor(shape:(3,), device=cpu, dtype=torch.int64),
 6)

---

source

TSClassification

 TSClassification (vocab=None, sort=True)

Vectorized, reversible transform of category string to vocab id

---

source

ToInt

 ToInt (enc=None, dec=None, split_idx=None, order=None)

Transforms an object dtype to int

---

source

ToFloat

 ToFloat (enc=None, dec=None, split_idx=None, order=None)

Transforms an object dtype to float (vectorized)

a = np.random.randint(0, 2, 10)
b = np.array(['1', '2', '3'])
c = np.array(['1.0', '2.0', '3.0'])
t = torch.randint(0, 2, (10, ))
test_eq(ToFloat()(a).dtype, 'float32')
test_eq(ToFloat()(b).dtype, 'float32')
test_eq(ToFloat()(c).dtype, 'float32')
test_eq(ToFloat()(t).dtype, torch.float32)

a = np.random.rand(10)*10
b = np.array(['1.0', '2.0', '3.0'])
t = torch.rand(10)*10
test_eq(ToInt()(a).dtype, 'int64')
test_eq(ToInt()(b).dtype, 'int64')
test_eq(ToInt()(t).dtype, torch.long)

t = TSClassification()
t.setup(y_on_disk[splits[0]])
y_encoded = t(y_on_disk)
print(y_encoded)
test_eq(t.decodes(y_encoded), y_on_disk)

TensorCategory([0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3,
                3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1,
                1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3])

y_multi= np.random.randint(0,3,20)
y_multi = np.asarray(alphabet[y_multi]).reshape(4,5)
tfm = TSClassification()
tfm.setup(y_multi)
enc_y_multi = tfm(y_multi)
test_eq(y_multi, tfm.decode(enc_y_multi))
enc_y_multi

TensorCategory([[2, 1, 0, 0, 1],
                [0, 0, 2, 1, 2],
                [0, 2, 0, 0, 0],
                [1, 1, 0, 0, 1]])

---

source

TSMultiLabelClassification

 TSMultiLabelClassification (c=None, vocab=None, add_na=False, sort=True)

Reversible combined transform of multi-category strings to one-hot encoded vocab id

---

source

TSTensorBlock

 TSTensorBlock (type_tfms=None, item_tfms=None, batch_tfms=None,
                dl_type=None, dls_kwargs=None)

Initialize self. See help(type(self)) for accurate signature.

---

source

NumpyTensorBlock

 NumpyTensorBlock (type_tfms=None, item_tfms=None, batch_tfms=None,
                   dl_type=None, dls_kwargs=None)

Initialize self. See help(type(self)) for accurate signature.

test_eq(NumpyTensorBlock().item_tfms[0].__name__, 'ToNumpyTensor')
test_eq(TSTensorBlock().item_tfms[0].__name__, 'ToTSTensor')

---

source

TSDataset

 TSDataset (X, y=None, split=None, sel_vars=None, sel_steps=None,
            types=None, dtype=None, device=None)

Initialize self. See help(type(self)) for accurate signature.

---

source

NumpyDataset

 NumpyDataset (X, y=None, types=None)

Initialize self. See help(type(self)) for accurate signature.

---

source

TorchDataset

 TorchDataset (X, y=None)

Initialize self. See help(type(self)) for accurate signature.

a = np.random.rand(5,6,7)
b = np.random.rand(5)
ds = NumpyDataset(a,b)
xb, yb = ds[[0,4]]
test_eq(xb.shape, (2,6,7))
test_eq(yb.shape, (2,))

---

source

TSTfmdLists

 TSTfmdLists (items=None, *rest, use_list=False, match=None)

A Pipeline of tfms applied to a collection of items

	Type	Default	Details
items	list		Items to apply Transforms to
use_list	bool	None	Use list in L

---

source

NoTfmLists

 NoTfmLists (items=None, *rest, use_list=False, match=None)

A Pipeline of tfms applied to a collection of items

items = X_on_disk
tl = TfmdLists(items, tfms=None, splits=splits)
test_eq(len(tl), len(X_on_disk))
test_eq(len(tl.train), len(splits[0]))
test_eq(len(tl.valid), len(splits[1]))
test_eq(tl[[0,4,7]], X_on_disk[[0,4,7]])
test_eq(tl.train[[0,4,7]], X_on_disk[splits[0][0,4,7]])
test_eq(tl.valid[[0,4,7]], X_on_disk[splits[1][0,4,7]])
test_eq(tl[0], items[0])
test_eq(tl[[0,1]], items[[0,1]])
test_eq(tl.decode(tl[0]), tl[0])
test_eq((tl.split_idx, tl.train.split_idx, tl.valid.split_idx), (None, 0, 1))

items = X_on_disk
tl = TSTfmdLists(items, tfms=None, splits=splits)
test_eq(len(tl), len(X_on_disk))
test_eq(len(tl.train), len(splits[0]))
test_eq(len(tl.valid), len(splits[1]))
test_eq(tl[[0,4,7]], X_on_disk[[0,4,7]])
test_eq(tl.train[[0,4,7]], X_on_disk[splits[0][0,4,7]])
test_eq(tl.valid[[0,4,7]], X_on_disk[splits[1][0,4,7]])
test_eq(tl[0], items[0])
test_eq(tl[[0,1]], items[[0,1]])
test_eq(tl.decode(tl[0]), tl[0])
test_eq((tl.split_idx, tl.train.split_idx, tl.valid.split_idx), (None, 0, 1))

items = X_on_disk
ntl = NoTfmLists(items, splits=splits)
test_eq(len(ntl), len(X_on_disk))
test_eq(len(ntl.train), len(splits[0]))
test_eq(len(ntl.valid), len(splits[1]))
test_eq(ntl._splits, np.arange(len(X_on_disk)))
test_eq(ntl.train._splits, np.arange(len(splits[0])))
test_eq(ntl.valid._splits, np.arange(len(splits[0]), len(X_on_disk)))
print(ntl)
print(ntl.train)
print(ntl.valid)
test_eq(ntl[[0,4,7]], X_on_disk[[0,4,7]])
test_eq(ntl.train[[0,4,7]], X_on_disk[splits[0][0,4,7]])
test_eq(ntl.valid[[0,4,7]], X_on_disk[splits[1][0,4,7]])
test_eq(ntl[0], items[0])
test_eq(ntl[[0,1]], items[[0,1]])
test_eq(ntl[:], X_on_disk)
ntl[0].shape, stack(ntl[[0,1]]).shape
test_eq(ntl.decode(ntl[0]), ntl[0])
assert id(items) == id(ntl.items) == id(ntl.train.items) == id(ntl.valid.items)
test_eq((ntl.split_idx, ntl.train.split_idx, ntl.valid.split_idx), (None, 0, 1))

NoTfmLists: memmap(60, 1, 570)
NoTfmLists: memmap(30, 1, 570)
NoTfmLists: memmap(30, 1, 570)

subitems = X_on_disk
new_ntl = ntl._new(X_on_disk)
test_eq(new_ntl[:], X_on_disk)

idxs = random_choice(len(X_on_disk), 10, False)
new_ntl = ntl._new(X_on_disk[idxs])
test_eq(new_ntl[:], X_on_disk[idxs])

idxs = random_choice(len(X_on_disk), 10, False)
new_ntl = ntl.valid._new(X_on_disk[idxs])
test_eq(new_ntl[:], X_on_disk[idxs])

---

source

tscoll_repr

 tscoll_repr (c, max_n=10)

String repr of up to max_n items of (possibly lazy) collection c

---

source

NumpyDatasets

 NumpyDatasets (items:list=None, tfms:MutableSequence|Pipeline=None,
                tls:TfmdLists=None, n_inp:int=None, dl_type=None,
                use_list:bool=None, do_setup:bool=True,
                split_idx:int=None, train_setup:bool=True,
                splits:list=None, types=None, verbose:bool=False)

A dataset that creates tuples from X (and y) and applies tfms of type item_tfms

	Type	Default	Details
items	NoneType	None
tfms	collections.abc.MutableSequence | fasttransform.transform.Pipeline		Transform(s) or Pipeline to apply
tls	NoneType	None
n_inp	NoneType	None
dl_type	TfmdDL	None	Type of DataLoader
use_list	bool	None	Use list in L
do_setup	bool	True	Call setup() for Transform
split_idx	int	None	Apply Transform(s) to training or validation set. 0 for training set and 1 for validation set
train_setup	bool	True	Apply Transform(s) only on training DataLoader
splits	list	None	Indices for training and validation sets
types	NoneType	None	Types of data in items
verbose	bool	False	Print verbose output

---

source

TSDatasets

 TSDatasets (items:list=None, tfms:MutableSequence|Pipeline=None,
             tls:TfmdLists=None, n_inp:int=None, dl_type=None,
             use_list:bool=None, do_setup:bool=True, split_idx:int=None,
             train_setup:bool=True, splits:list=None, types=None,
             verbose:bool=False)

A dataset that creates tuples from X (and optionally y) and applies item_tfms

	Type	Default	Details
items	NoneType	None
tfms	collections.abc.MutableSequence | fasttransform.transform.Pipeline		Transform(s) or Pipeline to apply
tls	NoneType	None
n_inp	NoneType	None
dl_type	TfmdDL	None	Type of DataLoader
use_list	bool	None	Use list in L
do_setup	bool	True	Call setup() for Transform
split_idx	int	None	Apply Transform(s) to training or validation set. 0 for training set and 1 for validation set
train_setup	bool	True	Apply Transform(s) only on training DataLoader
splits	list	None	Indices for training and validation sets
types	NoneType	None	Types of data in items
verbose	bool	False	Print verbose output

dsets = TSDatasets(X_on_disk, y_on_disk, splits=splits, tfms=[None, TSClassification()], inplace=True)
i = random_choice(len(splits[0]), 10, False).tolist()
test_eq(dsets.subset(i), dsets.train.subset(i))
dsets.valid.subset(i)
dsets.valid.subset(i)[[0,6,8]]
test_eq(dsets.subset(i)[[0,6,8]], dsets.train.subset(i)[[0,6,8]])
dsets.subset([0,7,3])
dsets.subset(i), dsets.train.subset(i), dsets.valid.subset(i)

((#10) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1))] ...],
 (#10) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1))] ...],
 (#10) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(2)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(3)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1))] ...])

tfms = [None, TSClassification()]
dsets = TSDatasets(X_on_disk, y_on_disk, splits=splits, tfms=tfms, inplace=False)
assert id(X_on_disk) == id(dsets.ptls[0].items) == id(dsets.train.ptls[0].items) == id(dsets.valid.ptls[0].items)

tfms = None
dsets = TSDatasets(X_on_disk, splits=splits, tfms=tfms, inplace=False)
assert id(X_on_disk) == id(dsets.ptls[0].items) == id(dsets.train.ptls[0].items) == id(dsets.valid.ptls[0].items)

---

source

TSDatasets.add_unlabeled

 TSDatasets.add_unlabeled (X, inplace=True)

---

source

TSDatasets.add_test

 TSDatasets.add_test (X, y=None, inplace=True)

---

source

TSDatasets.add_dataset

 TSDatasets.add_dataset (X, y=None, inplace=True)

---

source

NumpyDatasets.add_unlabeled

 NumpyDatasets.add_unlabeled (X, inplace=True)

---

source

NumpyDatasets.add_test

 NumpyDatasets.add_test (X, y=None, inplace=True)

---

source

NumpyDatasets.add_dataset

 NumpyDatasets.add_dataset (X, y=None, inplace=True)

---

source

add_ds

 add_ds (dsets, X, y=None, inplace=True)

Create test datasets from X (and y) using validation transforms of dsets

dsets = TSDatasets(X_on_disk, y_on_disk, splits=splits, tfms=[None, TSClassification()], inplace=True)
print(dsets.train[0][0].shape, dsets.train[[0,1]][0].shape)
print(dsets.split_idx, dsets.train.split_idx, dsets.valid.split_idx)
print(dsets.new_empty())
dsets

torch.Size([1, 570]) torch.Size([2, 1, 570])
None 0 1
(#0) []

(#60) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory(1))] ...]

dsets = TSDatasets(X_on_disk, y_on_disk, splits=splits, tfms=[None, TSClassification()], inplace=False)
print(dsets.train[0][0].shape, dsets.train[[0,1]][0].shape)
print(dsets.split_idx, dsets.train.split_idx, dsets.valid.split_idx)
print(dsets.new_empty())
dsets

torch.Size([1, 570]) torch.Size([2, 1, 570])
None 0 1
(#0) []

(#60) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([0])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([0])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([0])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([0])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([0])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([1])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([1])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([1])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([1])), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), TensorCategory([1]))] ...]

dsets = TSDatasets(X_on_disk, y_on_disk, tfms=[None, TSClassification()], splits=splits, inplace=True)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])

idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])
test_eq(dsets.train[idxs][1].numpy(), y_array[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])
test_eq(dsets.valid[idxs][1].numpy(), y_array[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, y_on_disk, tfms=[None, TSClassification()], splits=splits, inplace=False)
assert id(X_on_disk) == id(dsets.tls[0].items) == id(dsets.ptls[0].items)
assert id(X_on_disk) == id(dsets.train.tls[0].items) == id(dsets.train.ptls[0].items)
assert id(X_on_disk) == id(dsets.valid.tls[0].items) == id(dsets.valid.ptls[0].items)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])


idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])
test_eq(dsets.train[idxs][1].numpy(), y_array[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])
test_eq(dsets.valid[idxs][1].numpy(), y_array[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, splits=splits, inplace=True)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])

idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, splits=splits, inplace=False)
assert np.shares_memory(X_on_disk, dsets.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.ptls[0].items)
assert np.shares_memory(X_on_disk, dsets.train.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.train.ptls[0].items)
assert np.shares_memory(X_on_disk, dsets.valid.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.valid.ptls[0].items)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])

idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits, inplace=True)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])

idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])
test_eq(dsets.train[idxs][1].numpy(), y_array[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])
test_eq(dsets.valid[idxs][1].numpy(), y_array[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits, inplace=False)
assert np.shares_memory(X_on_disk, dsets.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.ptls[0].items)
assert np.shares_memory(X_on_disk, dsets.train.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.train.ptls[0].items)
assert np.shares_memory(X_on_disk, dsets.valid.tls[0].items)
assert np.shares_memory(X_on_disk, dsets.valid.ptls[0].items)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])

idxs = random_choice(len(dsets.train), 10, False)
test_eq(dsets.train[idxs][0].numpy(), X_on_disk[splits[0][idxs]])
test_eq(dsets.train[idxs][1].numpy(), y_array[splits[0][idxs]])

idxs = random_choice(len(dsets.valid), 10, False)
test_eq(dsets.valid[idxs][0].numpy(), X_on_disk[splits[1][idxs]])
test_eq(dsets.valid[idxs][1].numpy(), y_array[splits[1][idxs]])

dsets = TSDatasets(X_on_disk, y_on_disk, tfms=[None, TSClassification()], splits=None, inplace=True)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])

dsets = TSDatasets(X_on_disk, y_on_disk, tfms=[None, TSClassification()], splits=None, inplace=False)
assert id(X_on_disk) == id(dsets.tls[0].items) == id(dsets.ptls[0].items)
assert id(X_on_disk) == id(dsets.train.tls[0].items) == id(dsets.train.ptls[0].items)

idxs = random_choice(len(dsets), 10, False)
test_eq(dsets[idxs][0].numpy(), X_on_disk[idxs])
test_eq(dsets[idxs][1].numpy(), y_array[idxs])

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits)
test_eq(dsets.train[0:10], dsets.add_dataset(X_on_disk[0:10], y_array[0:10])[:])
test_eq(dsets.train[0:10][0], dsets.add_dataset(X_on_disk[0:10])[:][0])

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits)
torch.save(dsets, 'export/dsets.pth')
del dsets
dsets = torch.load('export/dsets.pth', weights_only=False)
dsets

(#60) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1))] ...]

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits)
torch.save(dsets.train, 'export/dsets.pth')
del dsets
dsets = torch.load('export/dsets.pth', weights_only=False)
dsets

(#30) [(TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(0)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1)), (TSTensor(vars:1, len:570, device=cpu, dtype=torch.float32), tensor(1))] ...]

dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits)
test_eq(len(dsets.train), len(X_train))
dsets = TSDatasets(X_on_disk, y_array, tfms=None, splits=splits)
test_eq(len(dsets.train), len(X_train))
dsets = TSDatasets(X_on_disk, y_array, tfms=[add(1), TSCategorize()], splits=splits)
test_eq(len(dsets.train), len(X_train))
# test_eq(dsets.train[0][0].data, tensor(X_train[0] + 1))
test_eq(dsets.train[0][1].item(), y_tensor[0])

dsets = TSDatasets(X_on_disk, y_on_disk, tfms=[None, TSCategorize()], splits=splits)
test_eq(len(dsets.add_test(X_train, y_train)), len(X_train))
test_eq(len(dsets.add_unlabeled(X_train)), len(X_train))

X_tensor = torch.randn(100, 4, 50)
y_tensor = torch.randint(0, 2, size=(len(X_tensor),))
tensor_splits = (np.arange(80), np.arange(80, 100))
dsets = TSDatasets(X_tensor, y_tensor, tfms=[None, TSClassification()], splits=tensor_splits)
test_eq(type(dsets[0][0]), TSTensor)

---

source

TSDataLoader

 TSDataLoader (dataset, bs=64, shuffle=False, drop_last=False,
               num_workers=0, verbose=False, do_setup=True, vocab=None,
               sort=False, weights=None, partial_n=None, sampler=None,
               pin_memory=False, timeout=0, batch_size=None, indexed=None,
               n=None, device=None, persistent_workers=False,
               pin_memory_device='', wif=None, before_iter=None,
               after_item=None, before_batch=None, after_batch=None,
               after_iter=None, create_batches=None, create_item=None,
               create_batch=None, retain=None, get_idxs=None, sample=None,
               shuffle_fn=None, do_batch=None)

Transformed DataLoader

	Type	Default	Details
dataset			Map- or iterable-style dataset from which to load the data
bs	int	64	Size of batch
shuffle	bool	False	Whether to shuffle data
drop_last	bool	False
num_workers	int	None	Number of CPU cores to use in parallel (default: All available up to 16)
verbose	bool	False	Whether to print verbose logs
do_setup	bool	True	Whether to run setup() for batch transform(s)
vocab	NoneType	None
sort	bool	False
weights	NoneType	None
partial_n	NoneType	None
sampler	NoneType	None
pin_memory	bool	False
timeout	int	0
batch_size	NoneType	None
indexed	NoneType	None
n	NoneType	None
device	NoneType	None
persistent_workers	bool	False
pin_memory_device	str
wif	NoneType	None
before_iter	NoneType	None
after_item	NoneType	None
before_batch	NoneType	None
after_batch	NoneType	None
after_iter	NoneType	None
create_batches	NoneType	None
create_item	NoneType	None
create_batch	NoneType	None
retain	NoneType	None
get_idxs	NoneType	None
sample	NoneType	None
shuffle_fn	NoneType	None
do_batch	NoneType	None

---

source

NumpyDataLoader

 NumpyDataLoader (dataset, bs=64, shuffle=False, drop_last=False,
                  num_workers=0, verbose=False, do_setup=True, vocab=None,
                  sort=False, weights=None, partial_n=None, sampler=None,
                  pin_memory=False, timeout=0, batch_size=None,
                  indexed=None, n=None, device=None,
                  persistent_workers=False, pin_memory_device='',
                  wif=None, before_iter=None, after_item=None,
                  before_batch=None, after_batch=None, after_iter=None,
                  create_batches=None, create_item=None,
                  create_batch=None, retain=None, get_idxs=None,
                  sample=None, shuffle_fn=None, do_batch=None)

Transformed DataLoader

	Type	Default	Details
dataset			Map- or iterable-style dataset from which to load the data
bs	int	64	Size of batch
shuffle	bool	False	Whether to shuffle data
drop_last	bool	False
num_workers	int	None	Number of CPU cores to use in parallel (default: All available up to 16)
verbose	bool	False	Whether to print verbose logs
do_setup	bool	True	Whether to run setup() for batch transform(s)
vocab	NoneType	None
sort	bool	False
weights	NoneType	None
partial_n	NoneType	None
sampler	NoneType	None
pin_memory	bool	False
timeout	int	0
batch_size	NoneType	None
indexed	NoneType	None
n	NoneType	None
device	NoneType	None
persistent_workers	bool	False
pin_memory_device	str
wif	NoneType	None
before_iter	NoneType	None
after_item	NoneType	None
before_batch	NoneType	None
after_batch	NoneType	None
after_iter	NoneType	None
create_batches	NoneType	None
create_item	NoneType	None
create_batch	NoneType	None
retain	NoneType	None
get_idxs	NoneType	None
sample	NoneType	None
shuffle_fn	NoneType	None
do_batch	NoneType	None

---

source

TSDataLoaders

 TSDataLoaders (*loaders, path='.', device=None)

Basic wrapper around several DataLoaders.

---

source

NumpyDataLoaders

 NumpyDataLoaders (*loaders, path='.', device=None)

Basic wrapper around several DataLoaders.

---

source

StratifiedSampler

 StratifiedSampler (y, bs:int=64, shuffle:bool=False,
                    drop_last:bool=False)

Sampler where batches preserve the percentage of samples for each class

	Type	Default	Details
y			The target variable for supervised learning problems. Stratification is done based on the y labels.
bs	int	64	Batch size
shuffle	bool	False	Flag to shuffle each class’s samples before splitting into batches.
drop_last	bool	False	Flag to drop the last incomplete batch.

a = np.concatenate([np.zeros(90), np.ones(10)])
sampler = StratifiedSampler(a, bs=32, shuffle=True, drop_last=True)
idxs = np.array(list(iter(sampler)))
print(idxs[:32])
print(a[idxs][:32])
test_eq(a[idxs][:32].mean(), .1)

[[ 1  2  5  8 10 14 16 17 26 27 31 33 34 35 38 39 50 51 53 54 55 56 58 62
  68 69 74 76 77 86 90 93 95 97  4  6  7  9 13 15 18 20 23 24 28 29 30 37
  41 43 44 45 60 64 70 71 75 78 79 80 81 82 84 88 92 94 98  0  3 11 12 19
  21 22 25 32 36 40 42 46 47 48 49 52 57 59 61 63 65 66 67 72 73 83 85 87
  89 91 96 99]]
[[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
  0. 0. 0. 0. 0. 0. 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
  0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0. 0. 0. 0. 0.
  0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
  0. 1. 1. 1.]]

---

source

get_c

 get_c (dls)

---

source

get_best_dls_params

 get_best_dls_params (dls, n_iters=10, num_workers=[0, 1, 2, 4, 8],
                      pin_memory=[True, False], prefetch_factor=[2, 4, 8],
                      return_best=True, verbose=True)

---

source

get_best_dl_params

 get_best_dl_params (dl, n_iters=10, num_workers=[0, 1, 2, 4, 8],
                     pin_memory=[True, False], prefetch_factor=[2, 4, 8],
                     return_best=True, verbose=True)

---

source

get_ts_dls

 get_ts_dls (X, y=None, splits=None, sel_vars=None, sel_steps=None,
             tfms=None, inplace=True, path='.', bs=64, batch_tfms=None,
             num_workers=0, device=None, shuffle_train=True,
             drop_last=True, weights=None, partial_n=None, sampler=None,
             sort=False, **kwargs)

# Tests
a = np.arange(10)

for s in [None, np.arange(10), np.arange(10).tolist(), L(np.arange(10).tolist()), (np.arange(10).tolist(), None), (np.arange(10).tolist(), L())]:
    test_eq(_check_splits(a, s), (L(np.arange(10).tolist()), L()))

---

source

get_subset_dl

 get_subset_dl (dl, idxs)

---

source

get_ts_dl

 get_ts_dl (X, y=None, split=None, sel_vars=None, sel_steps=None,
            tfms=None, inplace=True, path='.', bs=64, batch_tfms=None,
            num_workers=0, device=None, shuffle_train=True,
            drop_last=True, weights=None, partial_n=None, sampler=None,
            sort=False, **kwargs)

X, y, splits = get_UCR_data(dsid, on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=8)
dls = get_best_dls_params(dls, prefetch_factor=[2, 4, 8, 16])

Dataloader 0

   num_workers:  0  pin_memory: True   prefetch_factor:  2  -  time:    0.834 ms/iter
   num_workers:  0  pin_memory: False  prefetch_factor:  2  -  time:    0.472 ms/iter

   best dl params:
       best num_workers    : 0
       best pin_memory     : False
       best prefetch_factor: 2
       return_best         : True



Dataloader 1

   num_workers:  0  pin_memory: True   prefetch_factor:  2  -  time:    0.125 ms/iter
   num_workers:  0  pin_memory: False  prefetch_factor:  2  -  time:    0.090 ms/iter

   best dl params:
       best num_workers    : 0
       best pin_memory     : False
       best prefetch_factor: 2
       return_best         : True

y_int = np.random.randint(0, 4, size=len(X))
dls = get_ts_dls(X, y_int, splits=splits, bs=8)
test_eq(hasattr(dls, "vocab"), False)

dls = get_ts_dls(X, y_int, splits=splits, bs=8, vocab=[0,1,2,3])
test_eq(dls.vocab, [0,1,2,3])
test_eq(dls.c, 4)
test_eq(dls.cat, True)

X, y, splits = get_UCR_data(dsid, on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=8)
b=first(dls.train)
dls.decode(b)
test_eq(X.shape[1], dls.vars)
test_eq(X.shape[-1], dls.len)

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=64, inplace=True)

idxs = random_choice(len(dls.valid_ds), 10, False)
new_dl = get_subset_dl(dls.train, idxs)

idxs = random_choice(len(dls.valid_ds), 10, False)
new_dl = get_subset_dl(dls.valid, idxs)
test_eq(new_dl.one_batch()[0].cpu().numpy(), X[splits[1][idxs]])

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
weights = np.random.rand(len(X))
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=64, inplace=True, weights=weights)
weights2 = weights[splits[0]] / weights[splits[0]].sum()
test_eq(dls.train.weights, weights2)
test_eq(dls.valid.weights, None)

partial_n = 12
X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, splits=splits, tfms=[None, TSClassification()], bs=64, inplace=True, partial_n=partial_n)
test_eq(len(dls.train.one_batch()[0]), partial_n)

partial_n = .1
X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], bs=64, inplace=True, partial_n=partial_n)
test_eq(len(dls.train.one_batch()[0]), int(round(len(dls.train.dataset) * partial_n)))

You’ll now be able to pass a sampler to a tsai dataloader.

You should use a sampler for the train set and a sampler for the validation set. You’ll need to pass an object with the same length as each dataset. For example, the splits like in the case below.

⚠️ Remember to set shuffle=False when using a sampler since they a mutually exclusive. This means that when you use a sampler, you always need to set the shuffle in the dataloader to False. The sampler will control whether the indices are shuffled or not (you can set shuffle to True or False in the sampler).

drop_last is managed in the dataloder though.

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
train_sampler = torch.utils.data.sampler.RandomSampler(splits[0])
valid_sampler = torch.utils.data.sampler.SequentialSampler(splits[1])
dls = get_ts_dls(X, y, splits=splits, tfms=[None, TSClassification()], bs=8, inplace=True,
                 shuffle=False, drop_last=True, sampler=[train_sampler, valid_sampler])
print('train')
for _ in dls.train:
    print(dls.train.idxs)
print('valid')
for _ in dls.valid:
    print(dls.valid.idxs)

train
[np.int64(16), np.int64(5), np.int64(29), np.int64(23), np.int64(17), np.int64(3), np.int64(13), np.int64(26)]
[np.int64(24), np.int64(22), np.int64(11), np.int64(28), np.int64(10), np.int64(25), np.int64(27), np.int64(4)]
[np.int64(19), np.int64(0), np.int64(8), np.int64(18), np.int64(1), np.int64(20), np.int64(9), np.int64(7)]
valid
[np.int64(0), np.int64(1), np.int64(2), np.int64(3), np.int64(4), np.int64(5), np.int64(6), np.int64(7)]
[np.int64(8), np.int64(9), np.int64(10), np.int64(11), np.int64(12), np.int64(13), np.int64(14), np.int64(15)]
[np.int64(16), np.int64(17), np.int64(18), np.int64(19), np.int64(20), np.int64(21), np.int64(22), np.int64(23)]
[np.int64(24), np.int64(25), np.int64(26), np.int64(27), np.int64(28), np.int64(29)]

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
train_sampler = torch.utils.data.sampler.SequentialSampler(splits[0])
valid_sampler = torch.utils.data.sampler.SequentialSampler(splits[1])
dls = get_ts_dls(X, y, splits=splits, tfms=[None, TSClassification()], bs=64, inplace=True,
                 shuffle=False, sampler=[train_sampler, valid_sampler])
test_eq(dls.get_idxs(), np.arange(len(splits[0])))
test_eq(dls.train.get_idxs(), np.arange(len(splits[0])))
test_eq(dls.valid.get_idxs(), np.arange(len(splits[1])))
xb = dls.valid.one_batch()[0].cpu().numpy()
test_close(xb, X[dls.valid.split_idxs])

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
train_sampler = torch.utils.data.sampler.RandomSampler(splits[0])
valid_sampler = torch.utils.data.sampler.SequentialSampler(splits[0])
dls = get_ts_dls(X, y, splits=splits, tfms=[None, TSClassification()], bs=32, inplace=True,
                 shuffle=False, drop_last=True, sampler=[train_sampler, valid_sampler])
test_ne(dls.train.get_idxs(), np.arange(len(splits[0])))
test_eq(np.sort(dls.train.get_idxs()), np.arange(len(splits[0])))
test_eq(dls.valid.get_idxs(), np.arange(len(splits[1])))

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=64, inplace=False)

idxs = random_choice(len(dls.valid_ds), 10, False)
new_dl = get_subset_dl(dls.train, idxs)

idxs = random_choice(len(dls.valid_ds), 10, False)
new_dl = get_subset_dl(dls.valid, idxs)
test_eq(new_dl.one_batch()[0].cpu().numpy(), X[splits[1][idxs]])

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSClassification()], splits=splits, bs=8)
b = dls.one_batch()
input_idxs = dls.input_idxs
test_eq(b[0].cpu().numpy(), X[input_idxs])
b = dls.train.one_batch()
input_idxs = dls.train.input_idxs
test_eq(b[0].cpu().numpy(), X[input_idxs])
assert max(input_idxs) < len(splits[0])
b = dls.valid.one_batch()
input_idxs = dls.valid.input_idxs
test_eq(b[0].cpu().numpy(), X[input_idxs])
assert min(input_idxs) >= len(splits[0])

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSCategorize()], splits=splits, bs=8)
b=first(dls.train)
dls.decode(b)
test_eq(X.shape[1], dls.vars)
test_eq(X.shape[-1], dls.len)

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSCategorize()], splits=splits, bs=8, weights=np.random.randint(0, 3, len(y)))
b=first(dls.train)
dls.decode(b)
test_eq(X.shape[1], dls.vars)
test_eq(X.shape[-1], dls.len)

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dsets = TSDatasets(X, y, tfms=[None, TSCategorize()], splits=splits)
ts_dls = TSDataLoaders.from_dsets(dsets.train, dsets.valid, device=default_device(), bs=4)
torch.save(ts_dls, 'export/ts_dls.pth')
del ts_dls
ts_dls = torch.load('export/ts_dls.pth', weights_only=False)
for xb,yb in ts_dls.train:
    test_eq(tensor(X[ts_dls.train.idxs]), xb.cpu())

X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSCategorize()], splits=splits, bs=4)
for xb,yb in dls.train:
    test_eq(xb.cpu().numpy(), X[dls.train.input_idxs])
for xb,yb in dls.valid:
    test_eq(xb.cpu().numpy(), X[dls.valid.input_idxs])

test_eq((ts_dls.train.shuffle, ts_dls.valid.shuffle, ts_dls.train.drop_last, ts_dls.valid.drop_last), (True, False, True, False))

dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSCategorize()], splits=splits, bs=8, num_workers=0)
xb, yb = first(dls.train)
test_eq(tensor(X[dls.train.idxs]), xb.cpu())

test_eq((dls.train.shuffle, dls.valid.shuffle, dls.train.drop_last, dls.valid.drop_last), (True, False, True, False))

# multiclass
dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, on_disk=True, split_data=False)
dls = get_ts_dls(X, y, tfms=[None, TSCategorize()], splits=splits, inplace=True)
dls.show_dist()
dls.train.show_dist()
xb,yb = first(dls.train)
test_eq((dls.cat, dls.c), (True, 4))
test_ne(dls.cws.cpu().numpy(), None)
dls.decoder((xb, ))
dls.decoder((xb[0], ))
dls.decoder((xb, yb))
dls.decoder((xb[0], yb[0]))
dls.decoder(yb)
dls.decoder(yb[0])

'2'

new_dl = dls.new_dl(X)
first(new_dl)

(TSTensor(samples:60, vars:1, len:570, device=mps:0, dtype=torch.float32),)

new_dl = dls.new_dl(X, y=y)
first(new_dl)

(TSTensor(samples:60, vars:1, len:570, device=mps:0, dtype=torch.float32),
 TensorCategory([3, 3, 0, 0, 2, 3, 1, 0, 0, 3, 2, 2, 2, 3, 3, 1, 1, 2, 1, 3, 0,
                 1, 1, 3, 3, 3, 1, 3, 1, 3, 3, 3, 3, 3, 0, 1, 3, 0, 0, 1, 3, 1,
                 1, 1, 3, 3, 3, 3, 3, 1, 1, 0, 2, 2, 3, 3, 2, 1, 0, 1],
                device='mps:0'))

dls.train.dataset.split_idxs, dls.train.dataset.splits, dls.valid.split_idxs

(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
        17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29], dtype=int8),
 (#30) [np.int64(0),np.int64(1),np.int64(2),np.int64(3),np.int64(4),np.int64(5),np.int64(6),np.int64(7),np.int64(8),np.int64(9),np.int64(10),np.int64(11),np.int64(12),np.int64(13),np.int64(14),np.int64(15),np.int64(16),np.int64(17),np.int64(18),np.int64(19)...],
 array([30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46,
        47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59], dtype=int8))

# 2d input array and tfms == None return a NoTfmLists object
X, y, splits = get_UCR_data('OliveOil', on_disk=False, split_data=False)
X = X[:, 0]
tfms=[None, TSCategorize()]
dls = get_ts_dls(X, y, splits=splits, tfms=tfms, bs=8)
test_eq(1, dls.vars)
test_eq(X.shape[-1], dls.len)
test_eq(type(dls.tls[0]).__name__, 'NoTfmLists')
dls = get_ts_dls(X, y, splits=splits, tfms=tfms, bs=8, inplace=False)
test_eq(1, dls.vars)
test_eq(X.shape[-1], dls.len)
test_eq(type(dls.tls[0]).__name__, 'NoTfmLists')

# regression
dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, on_disk=True, split_data=False)
dls = get_ts_dls(X, np.random.rand(60, ), tfms=[None, ToNumpyTensor()], splits=splits)
dls.show_dist()
dls.train.show_dist()
xb,yb = first(dls.train)
dls.decoder((xb, ))
dls.decoder((xb[0], ))
dls.decoder((xb, yb))
dls.decoder((xb[0], yb[0]))
dls.decoder(yb)
dls.decoder(yb[0])
test_eq((dls.cat, dls.c), (False, 1))
test_eq(dls.cws, None)

# regression, multilabel
dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, on_disk=True, split_data=False)
dls = get_ts_dls(X, np.random.rand(60, 3) * 5, tfms=[None, ToNumpyTensor], splits=splits)
dls.show_dist()
dls.train.show_dist()
xb,yb = first(dls.train)
dls.decoder((xb, ))
dls.decoder((xb[0], ))
dls.decoder((xb, yb))
dls.decoder((xb[0], yb[0]))
dls.decoder(yb)
dls.decoder(yb[0])
test_eq((dls.cat, dls.c, dls.d),(False, 1, 3))
test_eq(dls.cws, None)

# multiclass, multilabel
dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, on_disk=True, split_data=False)
cm = {
    '1':'A',
    '2':['B', 'C'],
    '3':['B', 'D'] ,
    '4':'E',
    }
keys = cm.keys()
new_cm = {k:v for k,v in zip(keys, [listify(v) for v in cm.values()])}
y_multi = np.array([new_cm[yi] if yi in keys else listify(yi) for yi in y], dtype=object)
dls = get_ts_dls(X, y_multi, tfms=[None, TSMultiLabelClassification()], splits=splits)
dls.show_dist()
dls.train.show_dist()
xb,yb = first(dls.train)
dls.decoder((xb, ))
dls.decoder((xb[0], ))
dls.decoder((xb, yb))
dls.decoder((xb[0], yb[0]))
dls.decoder(yb)
dls.decoder(yb[0])
test_eq((dls.cat, dls.c), (True, 5))
test_ne(dls.cws.cpu().numpy(), None)

dsid = 'OliveOil'
X, y, splits = get_UCR_data(dsid, on_disk=True, split_data=False)
cm = {
    '1':'A',
    '2':['B', 'C'],
    '3':['B', 'D'] ,
    '4':'E',
    }
keys = cm.keys()
new_cm = {k:v for k,v in zip(keys, [listify(v) for v in cm.values()])}
y_multi = np.array([new_cm[yi] if yi in keys else listify(yi) for yi in y], dtype=object)
dls = get_ts_dls(X, y_multi, tfms=[None, TSMultiLabelClassification()], splits=splits)
test_eq(dls.new(X[0]).one_batch().shape, (1, 570))
test_eq(dls.new(X[:15]).one_batch().shape, (15, 1, 570))
test_eq(dls.train.new(X[0]).one_batch().shape, (1, 570))
test_eq(dls.valid.new(X[:15]).one_batch().shape, (15, 1, 570))

bs = 25
dsets = TSDatasets(X, y, tfms=[None, TSCategorize()], splits=splits)
dls   = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=[bs, bs*2], batch_tfms=add(1), num_workers=0)
xb,yb = dls.train.one_batch()
test_eq(xb.cpu().data, tensor(X_on_disk[splits[0]][dls.train.idxs]) + 1)

dsets = TSDatasets(X, y, tfms=[None, TSCategorize()], splits=splits)
dls   = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=[bs, bs*2])
xb,yb = dls.train.one_batch()
test_eq(xb.shape, (min(bs, len(splits[0])), X.shape[1], X.shape[-1]))
it = iter(dls.valid)
for xb,yb in it:
    test_close(xb.cpu(), TSTensor(X[splits[1]][dls.valid.idxs]))

bs = 64
dsets = TSDatasets(X, y, tfms=[add(1), TSCategorize()], splits=RandomSplitter(valid_pct=.3)(y_array))
dls = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=[bs, bs*2])
xb,yb = dls.train.one_batch()
test_eq(xb.shape, (min(bs, len(dsets.train)), X_on_disk.shape[1], X_on_disk.shape[-1]))
xb,yb = dls.valid.one_batch()
test_eq(xb.shape, (min(bs*2, len(dsets.valid)), X_on_disk.shape[1], X_on_disk.shape[-1]))

dsets = TSDatasets(X_on_disk, y_array, tfms=[None, TSCategorize()], splits=splits)
dls   = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=[32, 64])
for i in range(10):
    dl = dls.train if random.random() < .5 else dls.valid
    xb,yb = dl.one_batch()
    torch.equal(xb.cpu(), TSTensor(X_on_disk[dl.input_idxs]))

dsets = TSDatasets(X_on_disk, y_array, tfms=[None, TSCategorize()])
dls   = TSDataLoaders.from_dsets(dsets, bs=32)
for i in range(10):
    xb,yb = dls.one_batch()
    torch.equal(xb.cpu(), TSTensor(X_on_disk[dl.input_idxs]))

dsets = TSDatasets(X_on_disk, tfms=None)
dls   = TSDataLoaders.from_dsets(dsets, bs=32)
for i in range(10):
    xb = dls.one_batch()
    torch.equal(xb[0].cpu(), TSTensor(X_on_disk[dl.input_idxs]))

dsets = TSDatasets(X_on_disk, y_array, tfms=[None, TSCategorize()])
dls   = TSDataLoaders.from_dsets(dsets, bs=32)
test_eq(dls.split_idxs, L(np.arange(len(X_on_disk)).tolist()))

X, y, splits = get_UCR_data('NATOPS', return_split=False)
tfms  = [None, [TSCategorize()]]
dls = get_ts_dls(X, y, tfms=tfms, splits=splits, bs=[64, 128])
dls.show_batch()
dls.show_dist()

# test passing a list with categories instead of a numpy array
dsid = 'NATOPS'
bs = 64
X2, y2, splits2 = get_UCR_data(dsid, return_split=False)
vocab = sorted(set(y))
tfms = [None, [TSCategorize(vocab=vocab)]]
dsets = TSDatasets(X2, y2, tfms=tfms, splits=splits2)
dls = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=[bs, bs*2])
dls.train.one_batch()

(TSTensor(samples:64, vars:24, len:51, device=mps:0, dtype=torch.float32),
 TensorCategory([0, 2, 0, 3, 3, 0, 0, 4, 5, 1, 5, 5, 2, 0, 0, 4, 4, 5, 3, 2, 4,
                 5, 0, 3, 0, 0, 1, 4, 3, 3, 2, 5, 0, 3, 0, 5, 3, 1, 1, 0, 2, 4,
                 2, 5, 0, 4, 3, 3, 1, 2, 2, 0, 3, 1, 0, 2, 0, 3, 3, 2, 4, 5, 5,
                 2], device='mps:0'))

# MultiCategory
bs = 64
n_epochs = 100
tfms = [None, [MultiCategorize()]]
dsets = TSDatasets(X2, y2, tfms=tfms, splits=splits2)
dls = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=bs)
dls.train.one_batch()

(TSTensor(samples:64, vars:24, len:51, device=mps:0, dtype=torch.float32),
 TensorMultiCategory([[7, 0, 1],
                      [3, 0, 1],
                      [6, 0, 1],
                      [3, 0, 1],
                      [3, 0, 1],
                      [2, 0, 1],
                      [5, 0, 1],
                      [6, 0, 1],
                      [4, 0, 1],
                      [5, 0, 1],
                      [2, 0, 1],
                      [6, 0, 1],
                      [3, 0, 1],
                      [5, 0, 1],
                      [2, 0, 1],
                      [6, 0, 1],
                      [5, 0, 1],
                      [5, 0, 1],
                      [2, 0, 1],
                      [6, 0, 1],
                      [2, 0, 1],
                      [2, 0, 1],
                      [6, 0, 1],
                      [5, 0, 1],
                      [3, 0, 1],
                      [6, 0, 1],
                      [5, 0, 1],
                      [2, 0, 1],
                      [5, 0, 1],
                      [2, 0, 1],
                      [2, 0, 1],
                      [4, 0, 1],
                      [5, 0, 1],
                      [7, 0, 1],
                      [3, 0, 1],
                      [7, 0, 1],
                      [6, 0, 1],
                      [4, 0, 1],
                      [2, 0, 1],
                      [3, 0, 1],
                      [3, 0, 1],
                      [3, 0, 1],
                      [4, 0, 1],
                      [7, 0, 1],
                      [6, 0, 1],
                      [2, 0, 1],
                      [4, 0, 1],
                      [2, 0, 1],
                      [6, 0, 1],
                      [5, 0, 1],
                      [4, 0, 1],
                      [2, 0, 1],
                      [3, 0, 1],
                      [5, 0, 1],
                      [3, 0, 1],
                      [5, 0, 1],
                      [5, 0, 1],
                      [6, 0, 1],
                      [7, 0, 1],
                      [2, 0, 1],
                      [2, 0, 1],
                      [7, 0, 1],
                      [7, 0, 1],
                      [6, 0, 1]], device='mps:0'))

The combination of splits, sel_vars and sel_steps is very powerful, as it allows you to perform advanced indexing of the array-like X.

from tsai.data.validation import TSSplitter

X = np.arange(16*5*50).reshape(16,5,50)
y = alphabet[np.random.randint(0,3, 16)]
splits = TSSplitter(show_plot=False)(y)
tfms = [None, TSCategorize()]
batch_tfms = None
dls = get_ts_dls(X, y, splits=splits, sel_vars=[0, 1, 3], sel_steps=slice(-10, None), tfms=tfms, batch_tfms=batch_tfms)
xb,yb=dls.train.one_batch()
test_close(X[dls.input_idxs][:, [0, 1, 3]][...,slice(-10, None)], xb.cpu().numpy())
new_dl = dls.train.new_dl(X[:5], y[:5])
print(new_dl.one_batch())
new_empty_dl = dls.new_empty() # when exported
dl = new_empty_dl.new_dl(X[:10], y[:10], bs=64) # after export
dl.one_batch()

(TSTensor(samples:5, vars:3, len:10, device=mps:0, dtype=torch.int64), TensorCategory([0, 0, 0, 1, 0], device='mps:0'))

(TSTensor(samples:10, vars:3, len:10, device=mps:0, dtype=torch.int64),
 TensorCategory([1, 0, 1, 2, 0, 0, 0, 1, 0, 0], device='mps:0'))

---

source

get_dl_percent_per_epoch

 get_dl_percent_per_epoch (dl, model, n_batches=None)

---

source

get_time_per_batch

 get_time_per_batch (dl, model=None, n_batches=None)

X, y, splits = get_UCR_data('NATOPS', split_data=False)
tfms  = [None, [TSCategorize()]]
dls = get_ts_dls(X, y, tfms=tfms, splits=splits)
train_dl = dls.train
xb, _ = train_dl.one_batch()
model = nn.Linear(xb.shape[-1], 2).to(xb.device)
t = get_dl_percent_per_epoch(train_dl, model, n_batches=10)
print(t)

25.53%