New version

example.py

@@ -1,32 +1,24 @@
-from redisai import Client, Tensor, \
-    BlobTensor, DType, Device, Backend
+from redisai import Client, DType, Device, Backend
 import ml2rt
 
 client = Client()
-client.tensorset('x', Tensor(DType.float, [2], [2, 3]))
+client.tensorset('x', [2, 3], dtype=DType.float)
 t = client.tensorget('x')
 print(t.value)
 
 model = ml2rt.load_model('test/testdata/graph.pb')
-client.tensorset('a', Tensor.scalar(DType.float, 2, 3))
-client.tensorset('b', Tensor.scalar(DType.float, 12, 10))
+client.tensorset('a', (2, 3), dtype=DType.float)
+client.tensorset('b', (12, 10), dtype=DType.float)
 client.modelset('m', Backend.tf,
                 Device.cpu,
-                input=['a', 'b'],
-                output='mul',
+                inputs=['a', 'b'],
+                outputs='mul',
                 data=model)
 client.modelrun('m', ['a', 'b'], ['mul'])
 print(client.tensorget('mul').value)
 
 # Try with a script
 script = ml2rt.load_script('test/testdata/script.txt')
 client.scriptset('ket', Device.cpu, script)
-client.scriptrun('ket', 'bar', input=['a', 'b'], output='c')
+client.scriptrun('ket', 'bar', inputs=['a', 'b'], outputs='c')
 
-b1 = client.tensorget('c', as_type=BlobTensor)
-b2 = client.tensorget('c', as_type=BlobTensor)
-
-client.tensorset('d', BlobTensor(DType.float, b1.shape, b1, b2))
-
-tnp = b1.to_numpy()
-client.tensorset('e', tnp)

redisai/__init__.py

@@ -1,4 +1,3 @@
 from .version import __version__
 from .client import Client
-from .tensor import Tensor, BlobTensor
 from .constants import DType, Device, Backend

redisai/client.py

@@ -1,5 +1,6 @@
 from redis import StrictRedis
-from typing import Union, Any, AnyStr, ByteString, Sequence, Type
+from typing import Union, Any, AnyStr, ByteString, Sequence
+from .containers import Script, Model, Tensor
 
 try:
     import numpy as np
@@ -8,7 +9,7 @@
 
 from .constants import Backend, Device, DType
 from .utils import str_or_strsequence, to_string
-from .tensor import Tensor, BlobTensor
+from . import tensorize
 
 
 class Client(StrictRedis):
@@ -45,13 +46,12 @@ def modelset(self,
         args += [data]
         return self.execute_command(*args)
 
-    def modelget(self, name: AnyStr) -> dict:
+    def modelget(self, name: AnyStr) -> Model:
         rv = self.execute_command('AI.MODELGET', name)
-        return {
-            'backend': Backend(to_string(rv[0])),
-            'device': Device(to_string(rv[1])),
-            'data': rv[2]
-        }
+        return Model(
+            rv[2],
+            Device(to_string(rv[1])),
+            Backend(to_string(rv[0])))
 
     def modeldel(self, name: AnyStr) -> AnyStr:
         return self.execute_command('AI.MODELDEL', name)
@@ -68,71 +68,64 @@ def modelrun(self,
 
     def tensorset(self,
                   key: AnyStr,
-                  tensor: Union[Tensor, np.ndarray, list, tuple],
+                  tensor: Union[np.ndarray, list, tuple],
                   shape: Union[Sequence[int], None] = None,
                   dtype: Union[DType, None] = None) -> Any:
         """
         Set the values of the tensor on the server using the provided Tensor object
         :param key: The name of the tensor
-        :param tensor: a `Tensor` object
-        :param shape: Shape of the tensor
-        :param dtype: data type of the tensor. Required if input is a sequence of ints/floats
+        :param tensor: a `np.ndarray` object or python list or tuple
+        :param shape: Shape of the tensor. Required if `tensor` is list or tuple
+        :param dtype: data type of the tensor. Required if `tensor` is list or tuple
         """
-        # TODO: tensorset will not accept BlobTensor or Tensor object in the future.
-        # Keeping it in the current version for compatibility with the example repo
         if np and isinstance(tensor, np.ndarray):
-            tensor = BlobTensor.from_numpy(tensor)
+            tensor = tensorize.from_numpy(tensor)
+            args = ['AI.TENSORSET', key, tensor.dtype.value, *tensor.shape, tensor.argname, tensor.value]
         elif isinstance(tensor, (list, tuple)):
             if shape is None:
                 shape = (len(tensor),)
-            tensor = Tensor(dtype, shape, tensor)
-        args = ['AI.TENSORSET', key, tensor.type.value]
-        args += tensor.shape
-        args += [tensor.ARGNAME]
-        args += tensor.value
+            tensor = tensorize.from_sequence(tensor, shape, dtype)
+            args = ['AI.TENSORSET', key, tensor.dtype.value, *tensor.shape, tensor.argname, *tensor.value]
         return self.execute_command(*args)
 
     def tensorget(self,
-                  key: AnyStr, as_type: Type[Tensor] = None,
-                  meta_only: bool = False) -> Union[Tensor, BlobTensor]:
+                  key: AnyStr, as_numpy: bool = True,
+                  meta_only: bool = False) -> Union[Tensor, np.ndarray]:
         """
         Retrieve the value of a tensor from the server. By default it returns the numpy array
         but it can be controlled using `as_type` argument and `meta_only` argument.
         :param key: the name of the tensor
-        :param as_type: the resultant tensor type. Returns numpy array if None
+        :param as_numpy: Should it return data as numpy.ndarray.
+            Wraps with namedtuple if False. This flag also decides how to fetch the
+            value from RedisAI server and could have performance implications
         :param meta_only: if true, then the value is not retrieved,
             only the shape and the type
         :return: an instance of as_type
         """
-        # TODO; We might remove Tensor & BlobTensor in the future and `tensorget` will return
-        # python list or numpy arrays or a namedtuple
         if meta_only:
             argname = 'META'
-        elif as_type is None:
-            argname = BlobTensor.ARGNAME
+        elif as_numpy is True:
+            argname = 'BLOB'
         else:
-            argname = as_type.ARGNAME
+            argname = 'VALUES'
 
         res = self.execute_command('AI.TENSORGET', key, argname)
         dtype, shape = to_string(res[0]), res[1]
-        dt = DType.__members__[dtype.lower()]
         if meta_only:
-            return Tensor(dt, shape, [])
-        elif as_type is None:
-            return BlobTensor.from_resp(dt, shape, res[2]).to_numpy()
+            return tensorize.to_sequence([], shape, dtype)
+        if as_numpy is True:
+            return tensorize.to_numpy(res[2], shape, dtype)
         else:
-            return as_type.from_resp(dt, shape, res[2])
+            return tensorize.to_sequence(res[2], shape, dtype)
 
     def scriptset(self, name: AnyStr, device: Device, script: AnyStr) -> AnyStr:
         return self.execute_command('AI.SCRIPTSET', name, device.value, script)
 
-    def scriptget(self, name: AnyStr) -> dict:
+    def scriptget(self, name: AnyStr) -> Script:
         r = self.execute_command('AI.SCRIPTGET', name)
-        device = Device(to_string(r[0]))
-        return {
-            'device': device,
-            'script': to_string(r[1])
-        }
+        return Script(
+            to_string(r[1]),
+            Device(to_string(r[0])))
 
     def scriptdel(self, name):
         return self.execute_command('AI.SCRIPTDEL', name)

redisai/constants.py

@@ -10,6 +10,7 @@ class Backend(Enum):
     tf = 'TF'
     torch = 'TORCH'
     onnx = 'ONNX'
+    tflite = 'TFLITE'
 
 
 class DType(Enum):

redisai/containers.py

@@ -0,0 +1,5 @@
+from collections import namedtuple
+
+Tensor = namedtuple('Tensor', field_names=['value', 'shape', 'dtype', 'argname'])
+Script = namedtuple('Script', field_names=['script', 'device'])
+Model = namedtuple('Model', field_names=['data', 'device', 'backend'])

redisai/tensorize.py

@@ -0,0 +1,43 @@
+from typing import Union, ByteString, Sequence
+from .utils import convert_to_num
+from .constants import DType
+from .containers import Tensor
+try:
+    import numpy as np
+except (ImportError, ModuleNotFoundError):
+    np = None
+
+
+def from_numpy(tensor: np.ndarray) -> Tensor:
+    """ Convert the numpy input from user to `Tensor` """
+    dtype = DType.__members__[str(tensor.dtype)]
+    shape = tensor.shape
+    blob = bytes(tensor.data)
+    return Tensor(blob, shape, dtype, 'BLOB')
+
+
+def from_sequence(tensor: Sequence, shape: Union[list, tuple], dtype: DType) -> Tensor:
+    """ Convert the `list`/`tuple` input from user to `Tensor` """
+    return Tensor(tensor, shape, dtype, 'VALUES')
+
+
+def to_numpy(value: ByteString, shape: Union[list, tuple], dtype: DType) -> np.ndarray:
+    """ Convert `BLOB` result from RedisAI to `np.ndarray` """
+    dtype = DType.__members__[dtype.lower()].value
+    mm = {
+        'FLOAT': 'float32',
+        'DOUBLE': 'float64'
+    }
+    if dtype in mm:
+        dtype = mm[dtype]
+    else:
+        dtype = dtype.lower()
+    a = np.frombuffer(value, dtype=dtype)
+    return a.reshape(shape)
+
+
+def to_sequence(value: list, shape: list, dtype: DType) -> Tensor:
+    """ Convert `VALUES` result from RedisAI to `Tensor` """
+    dtype = DType.__members__[dtype.lower()]
+    convert_to_num(dtype, value)
+    return Tensor(value, tuple(shape), dtype, 'VALUES')

redisai/version.py

@@ -2,4 +2,4 @@
 # 1) we don't load dependencies by storing it in __init__.py
 # 2) we can import it in setup.py for the same reason
 # 3) we can import it into your module module
-__version__ = '0.4.1'
+__version__ = '0.5.0'

setup.py

@@ -1,4 +1,3 @@
-
 #!/usr/bin/env python
 from setuptools import setup, find_packages