cntk.contrib.crosstalkcaffe.unimodel.cntkinstance module¶

class ApiSetup[source]¶

Bases: object

Setup CNTK ops with given parameters

static batch_norm(cntk_layer, inputs)[source]¶

Setup batch normalization op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of batch normalization op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk batch normalization op
Return type:	`Function()`

static convolution(cntk_layer, inputs)[source]¶

Setup convolution op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of convolution op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk convolution op
Return type:	`Function()`

static dense(cntk_layer, inputs)[source]¶

Setup dense op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of dense op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk dense op
Return type:	`Function()`

static dropout(cntk_layer, inputs)[source]¶

Setup dropout op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of dropout op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk dropout op
Return type:	`Function()`

static lrn(cntk_layer, inputs)[source]¶

Setup lrn op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of lrn op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk lrn op
Return type:	`Function()`

static plus(cntk_layer, inputs)[source]¶

Setup plus op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of dense op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk dense op
Return type:	`Function()`

static pooling(cntk_layer, inputs)[source]¶

Setup pooling op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of pooling op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk pooling op
Return type:	`Function()`

static relu(cntk_layer, inputs)[source]¶

Setup ReLU op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of ReLU op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk ReLU op
Return type:	`Function()`

static softmax(cntk_layer, inputs)[source]¶

Setup softmax op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of softmax op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk softmax op
Return type:	`Function()`

static splice(cntk_layer, inputs)[source]¶

Setup splice op with given parameters

Parameters:	cntk_layer (`CntkLayersDefinition`) – the layer definition of splice op inputs (list) – a list contains all `Function` or `input`
Returns:	instaced cntk splice op
Return type:	`Function()`

class BlockApiSetup[source]¶

Bases: object

Implement some special requirement ops

static convolution(output, kernel, stride, pad, kernel_init, bias_init, group, dilation, name)[source]¶

Implement convolution ops

Parameters:	output (int) – the output channel size kernel (list) – the kernel size of filter, with format [width, height] stride (list) – the stride of convolution, with format [w_stride, h_stride] pad (bool) – auto padding or not kernel_init (np.array) – the tensor saving initialize values of filter bias_init (np.array) – the tensor saving initialize values of bias group (int) – the group size in the convolution dilation (list) – the dilation of convolution, with format [w_dilation, h_dilation] name (str) – the name of ops
Returns:	the function contains convolution ops
Return type:	`as_block()`

static linear(output_shape, input_shape, scale_init, bias_init, name)[source]¶

Implement linear ops, also known as full connection in Caffe

Parameters:	output_shape (tuple) – the output channel size input_shape (tuple) – the input channel size scale_init (np.array) – the tensor saving initialize values of scale bias_init (np.array) – the tensor saving initialize values of bias name (str) – the name of ops
Returns:	the function contains linear ops
Return type:	`as_block()`

static lrn(k, n, alpha, beta, name)[source]¶

Implement LRN ops

Parameters:	k (int) – the factor k in LRN n (int) – the normalization radius alpha (float) – alpha factor in LRN beta (float) – beta factor in LRN name (str) – the name of ops
Returns:	the function contains lrn ops
Return type:	`as_block()`

class CntkApiInstance(cntk_uni_model, global_conf)[source]¶

Bases: object

Instace CNTK ops and network

export_model()[source]¶

Save instanced CNTK model

Parameters:	None –
Returns:	None

get_functions()[source]¶

Return the functions of CNTK network

Parameters:	None –
Returns:	the instaced functions of CNTK
Return type:	list

get_model()[source]¶

Get instaced CNTK model

Parameters:	None –
Returns:	the output node of CNTK
Return type:	`Function()`

instance_functions(cntk_sorted_layers, cntk_layers)[source]¶

Instace all nodes into CNTK ops

Parameters:	cntk_sorted_layers (list) – the list contains the name of instaced layers with traversal order cntk_layers (dict) – the dict contains all layers definition
Returns:	None

instance_input(data_providers)[source]¶

Instace the inputs into CNTK variable

Parameters:	data_providers (list) – the list contains the definition of inputs
Returns:	None