Regenrated docs and pngs

.github/workflows/ci_test_all.yml

@@ -2,9 +2,9 @@ name: CI Test script
 
 on:
   push:
-    branches: [ main, development, experimental, test*, nml* ]
+    branches: [ main, development, experimental, test*, nml*, version* ]
   pull_request:
-    branches: [ main, development, experimental, test*, nml* ]
+    branches: [ main, development, experimental, test*, nml*, version* ]
 
 jobs:
 

README.md

@@ -13,6 +13,13 @@
 **Note: MDF is still in development! See the [open issues related to the specification](https://github.com/ModECI/MDF/issues?q=is%3Aissue+is%3Aopen+label%3Aspecification) or go [here](http://modeci.org/#contactPage) to get in contact regarding MDF.**
 *The MDF format was first proposed following a meeting organised at Princeton in July 2019 by Russ Poldrack of the Center for Reproducible Neuroscience (CRN) at Stanford and the [Brain Imaging Data Standard (BIDS)](https://bids.neuroimaging.io/) initiative. For more on the previous work in this area, see [here](https://github.com/OpenSourceBrain/PsyNeuLinkShowcase/tree/master/BIDS-MDF).*
 
+## Paper introducing MDF
+
+The background to the ModECI project, the motivation for developing the Model Description Format, and the initial Python implementation of the language have been described in a NeuroView article in the Neuron journal:
+
+*<b>Integrating model development across computational neuroscience, cognitive science, and machine learning</b>*
+Padraig Gleeson, Sharon Crook, David Turner, Katherine Mantel, Mayank Raunak, Ted Willke and Jonathan D. Cohen, April 25, 2023 DOI: [https://doi.org/10.1016/j.neuron.2023.03.037](https://doi.org/10.1016/j.neuron.2023.03.037)
+
 
 ## Overview
 

docs/MDF_function_specifications.json

@@ -686,15 +686,15 @@
         "expression_string": "onnx_ops.lppool(X, auto_pad, kernel_shape, p, pads, strides)"
     },
     "onnx::MatMul": {
-        "description": "\nMatrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html\n",
+        "description": "\nMatrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).\n",
         "arguments": [
             "A",
             "B"
         ],
         "expression_string": "onnx_ops.matmul(A, B)"
     },
     "onnx::MatMulInteger": {
-        "description": "\nMatrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.\nThe production MUST never overflow. The accumulation may overflow if and only if in 32 bits.\n",
+        "description": "\nMatrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).\nThe production MUST never overflow. The accumulation may overflow if and only if in 32 bits.\n",
         "arguments": [
             "A",
             "B",
@@ -711,7 +711,7 @@
         "expression_string": "onnx_ops.max(data_0)"
     },
     "onnx::MaxPool": {
-        "description": "\n MaxPool consumes an input tensor X and applies max pooling across\n the tensor according to kernel sizes, stride sizes, and pad lengths.\n max pooling consisting of computing the max on all values of a\n subset of the input tensor according to the kernel size and downsampling the\n data into the output tensor Y for further processing. The output spatial shape is calculated differently\n depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized.\n With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d):\n ```\n output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1)\n ```\n or\n ```\n output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1)\n ```\n if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored.\n\n `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled:\n ```\n VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i])\n SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])\n ```\n or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D):\n ```\n VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1\n SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1\n ```\n And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`:\n ```\n pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]\n ```\n The output of each pooling window is maximum number of elements exclude pad. \n ",
+        "description": "\n MaxPool consumes an input tensor X and applies max pooling across\n the tensor according to kernel sizes, stride sizes, and pad lengths.\n max pooling consisting of computing the max on all values of a\n subset of the input tensor according to the kernel size and downsampling the\n data into the output tensor Y for further processing. The output spatial shape is calculated differently\n depending on whether explicit padding is used, where pads is employed, or auto padding is used, where auto_pad is utilized.\n With explicit padding (https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html?highlight=maxpool#torch.nn.MaxPool2d):\n ```\n output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1)\n ```\n or\n ```\n output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1)\n ```\n if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`.\n\n `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled:\n ```\n VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i])\n SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])\n ```\n or when ceil_mode is disabled (https://www.tensorflow.org/api_docs/python/tf/keras/layers/AveragePooling2D):\n ```\n VALID: output_spatial_shape[i] = floor((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i]) + 1\n SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = floor((input_spatial_shape[i] - 1) / strides_spatial_shape[i]) + 1\n ```\n And pad shape will be following if `SAME_UPPER` or `SAME_LOWER`:\n ```\n pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]\n ```\n The output of each pooling window is maximum number of elements exclude pad. \n ",
         "arguments": [
             "X"
         ],
@@ -900,7 +900,7 @@
         "expression_string": "onnx_ops.qlinearconv(x, x_scale, x_zero_point, w, w_scale, w_zero_point, y_scale, y_zero_point, B, auto_pad, dilations, group, kernel_shape, pads, strides)"
     },
     "onnx::QLinearMatMul": {
-        "description": "\nMatrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.\nIt consumes two quantized input tensors, their scales and zero points, scale and zero point of output,\nand computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point).\nFor (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details.\nScale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor\n(per row for 'a' and per column for 'b'). Scalar refers to per tensor quantization whereas N-D refers to per row\nor per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be\nan M element vector [v_1, v_2, ..., v_M] for per row quantization and K element vector of shape [v_1, v_2, ..., v_K]\nfor per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may\nhave shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization.\nProduction must never overflow, and accumulation may overflow if and only if in 32 bits.\n",
+        "description": "\nMatrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).\nIt consumes two quantized input tensors, their scales and zero points, scale and zero point of output,\nand computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point).\nFor (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details.\nScale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor\n(per row for 'a' and per column for 'b'). Scalar refers to per tensor quantization whereas N-D refers to per row\nor per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be\nan M element vector [v_1, v_2, ..., v_M] for per row quantization and K element vector of shape [v_1, v_2, ..., v_K]\nfor per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may\nhave shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization.\nProduction must never overflow, and accumulation may overflow if and only if in 32 bits.\n",
         "arguments": [
             "a",
             "a_scale",

docs/MDF_function_specifications.md

@@ -1685,7 +1685,7 @@ Python version: `onnx_ops.lppool(X, auto_pad, kernel_shape, p, pads, strides)`
 
 ## MatMul
  <p><i>
-Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
+Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 </i></p>
 
 Python version: `onnx_ops.matmul(A, B)`
@@ -1695,7 +1695,7 @@ Python version: `onnx_ops.matmul(A, B)`
 
 ## MatMulInteger
  <p><i>
-Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
+Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
 </i></p>
 
@@ -1732,7 +1732,7 @@ Python version: `onnx_ops.max(data_0)`
  ```
  output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i] + 1)
  ```
- if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`. Sliding windows that would start in the right padded region are ignored.
+ if ceil_mode is enabled. `pad_shape[i]` is the sum of pads along axis `i`.
 
  `auto_pad` is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following when ceil_mode is enabled:
  ```
@@ -2237,7 +2237,7 @@ Python version: `onnx_ops.qlinearconv(x, x_scale, x_zero_point, w, w_scale, w_ze
 
 ## QLinearMatMul
  <p><i>
-Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
+Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 It consumes two quantized input tensors, their scales and zero points, scale and zero point of output,
 and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point).
 For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details.

docs/MDF_function_specifications.yaml

@@ -1617,7 +1617,7 @@ onnx::LpPool:
 onnx::MatMul:
     description: '
 
-        Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
+        Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 
         '
     arguments:
@@ -1627,7 +1627,7 @@ onnx::MatMul:
 onnx::MatMulInteger:
     description: '
 
-        Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
+        Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 
         The production MUST never overflow. The accumulation may overflow if and only
         if in 32 bits.
@@ -1668,8 +1668,7 @@ onnx::MaxPool:
         \ 1)\n ```\n or\n ```\n output_spatial_shape[i] = ceil((input_spatial_shape[i]\
         \ + pad_shape[i] - dilation[i] * (kernel_shape[i] - 1) - 1) / strides_spatial_shape[i]\
         \ + 1)\n ```\n if ceil_mode is enabled. `pad_shape[i]` is the sum of pads\
-        \ along axis `i`. Sliding windows that would start in the right padded region\
-        \ are ignored.\n\n `auto_pad` is a DEPRECATED attribute. If you are using\
+        \ along axis `i`.\n\n `auto_pad` is a DEPRECATED attribute. If you are using\
         \ them currently, the output spatial shape will be following when ceil_mode\
         \ is enabled:\n ```\n VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i]\
         \ - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i])\n\
@@ -2090,7 +2089,7 @@ onnx::QLinearConv:
 onnx::QLinearMatMul:
     description: '
 
-        Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
+        Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html).
 
         It consumes two quantized input tensors, their scales and zero points, scale
         and zero point of output,