Merge branch 'ml4ai:main' into main

.github/workflows/tests-and-docs.yml

@@ -81,14 +81,17 @@ jobs:
       working-directory: .
       run: |
         # retrieve latest model for img2mml component
-        curl -L https://artifacts.askem.lum.ai/skema/img2mml/models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt > skema/img2mml/trained_models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
+        pip install huggingface_hub
+        python scripts/retrieve_model_ci.py
+
+        # Install askem
         pip install ".[all]"
+
     # Install tree-sitter parser (for Python component unit tests)
     - name: Install tree-sitter parsers
       working-directory: .
       run: python skema/program_analysis/tree_sitter_parsers/build_parsers.py --ci --all
 
-
     # docs (API)
     # generate python docs using pdoc
     - name: "Create documentation for Python components (API docs)"

Dockerfile.skema-py

@@ -54,7 +54,7 @@ ENV PATH="/root/.cargo/bin:${PATH}"
 RUN pip install wheel
 RUN pip install six
 # Download ML model (~150MB)
-RUN curl -L https://artifacts.askem.lum.ai/skema/img2mml/models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt > skema/img2mml/trained_models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
+RUN pip install huggingface_hub && python scripts/retrieve_model_ci.py
 RUN tree /app
 #RUN pip install ".[all]"
 # exclude dependencies for docs

docs/dev/env.md

@@ -9,7 +9,7 @@ conda activate skema
 # Install tree-sitter parsers
 python skema/program_analysis/tree_sitter_parsers/build_parsers.py --all
 # download the checkpoint for the img2mml service
-curl -L https://artifacts.askem.lum.ai/skema/img2mml/models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt > skema/img2mml/trained_models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
+python scripts/retrieve_model.py
 # mathjax deps for img2mml
 (cd skema/img2mml/data_generation && npm install)
 ```

pyproject.toml

@@ -15,19 +15,19 @@ dependencies=[
     "numpy",
     "dill==0.3.7",
     "networkx==2.8.8",
-    "PyYAML",
+    "PyYAML==6.*",
     "tree-sitter==0.20.4",
     "neo4j==5.14.1",
     "requests",
-    "beautifulsoup4", # used to remove comments etc from pMML before sending to MORAE
+    "beautifulsoup4==4.12.*", # used to remove comments etc from pMML before sending to MORAE
     "typing_extensions", # see https://github.com/pydantic/pydantic/issues/5821#issuecomment-1559196859
     "fastapi~=0.100.0",
     "starlette",
     "httpx",
     "pydantic>=2.0.0",
     "uvicorn",
     "python-multipart",
-    "func_timeout"
+    "func_timeout==4.3.5"
 ]
 # The Python program analysis pipeline does not currently work with Python 3.9
 # or 3.10. This may change in the future.
@@ -57,7 +57,7 @@ isa = [
 ]
 
 # shared ML dependencies
-ml = ["torch==2.0.1", "torchvision==0.15.2", "beartype==0.15.0"]
+ml = ["torch==2.0.1", "torchvision==0.15.2", "beartype==0.15.0", "huggingface_hub"]
 
 # Im2MML dependencies. The img2mml service converts equation images to MathML.
 # See the skema/img2mml directory.
@@ -100,6 +100,7 @@ all = ["skema[core]", "skema[dev]", "skema[doc]", "skema[demo]", "skema[annotati
 "skema.rest" = "skema/rest"
 "skema.skema_py" = "skema/skema_py"
 "skema.utils" = "skema/utils"
+"skema.data" = "skema/data"
 
 # re-map skema/text_reading/python to skema.text_reading
 #"skema.text_reading" = "skema/text_reading/python"
@@ -110,7 +111,7 @@ all = ["skema[core]", "skema[dev]", "skema[doc]", "skema[demo]", "skema[annotati
 
 [tool.setuptools.package-data]
 # needed to ensure models are included in package/discoverable
-"*" = ["*.json", "vocab.txt", "*.pt", "*.png", "*.html", "*.yml", "*.yaml"]
+"*" = ["*.json", "*vocab.txt", "*.pt", "*.png", "*.html", "*.yml", "*.yaml", "*.zip"]
 
 [tool.setuptools.dynamic]
 readme = {file = ["README.md"], content-type = "text/markdown"}

scripts/retrieve_model_ci.py

@@ -0,0 +1,31 @@
+import os
+from pathlib import Path
+
+from huggingface_hub import hf_hub_download
+
+def retrieve_model(model_path=None) -> str:
+    """
+    Retrieve the img2mml model from the specified path or download it if not found.
+
+    Args:
+        model_path (str, optional): Path to the img2mml model file. Defaults to None.
+
+    Returns:
+        str: Path to the loaded model file.
+    """
+    cwd = Path(__file__).parents[0]
+    REPO_NAME = "lum-ai/img2mml"
+    MODEL_NAME = "cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt"
+    # If the model path is none or doesn't exist, the default model will be downloaded from server.
+    if model_path is None or not os.path.exists(model_path):
+        model_path = cwd / "trained_models" / MODEL_NAME
+
+        # Check if the model file already exists
+        if not os.path.exists(model_path):
+            # If the file doesn't exist, download it from the specified URL
+            print(f"Downloading the model checkpoint from HuggingFace...")
+            hf_hub_download(repo_id=REPO_NAME, filename=MODEL_NAME, local_dir=model_path.parent, local_dir_use_symlinks=False)
+
+    return str(model_path)
+
+retrieve_model()

skema/data/program_analysis/__init__.py

@@ -0,0 +1,3 @@
+from pathlib import Path
+
+MODEL_ZIP_ROOT_PATH = Path(__file__).parent

skema/img2mml/README.md

@@ -12,24 +12,23 @@ service was developed by Deepsana Shahi, Adarsh Pyarelal and Liang Zhang.
 
 The model itself is not checked into the repository, but you can get it from
 here:
-https://artifacts.askem.lum.ai/skema/img2mml/models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
+https://huggingface.co/lum-ai/img2mml/blob/main/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
 
 Place the model file in the `trained_models` directory.
 
-The curl command below should do the trick.
+The Python command below should do the trick.
 
 ```
-curl -L https://artifacts.askem.lum.ai/skema/img2mml/models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt > trained_models/cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt
+python ../../scripts/retrieve_model_ci.py
 ```
 
-If you have the checkpoint in the `trained_models` directory already and hope to update it, please run the above curl command that will replace the previous one.
+If you have the checkpoint in the `trained_models` directory already and hope to update it, please run the above Python command that will replace the previous one.
 
 To update the model name or path, please make the following modifications to support updating the img2mml service and the corresponding Docker operations:
 
 1. Modify the ENV variable of `SKEMA_IMG2MML_MODEL_PATH`.
 2. Update the path settings in the "retrieve latest model for img2mml component" section of `skema/.github/workflows/tests-and-docs.yml`.
-3. Adjust the curl command in the test_equation_reading section of `skema/.drone.yml` to download the checkpoint.
-4. Update the download checkpoint path in `skema/img2mml/README.md`.
+3. Update the download checkpoint path in `skema/img2mml/README.md`.
 
 These changes will ensure that the necessary files and paths are updated correctly.
 

skema/img2mml/api.py

@@ -13,6 +13,7 @@
 from PIL import Image
 from io import BytesIO
 
+from huggingface_hub import hf_hub_download
 
 def retrieve_model(model_path=None) -> str:
     """
@@ -25,7 +26,7 @@ def retrieve_model(model_path=None) -> str:
         str: Path to the loaded model file.
     """
     cwd = Path(__file__).parents[0]
-    MODEL_BASE_ADDRESS = "https://artifacts.askem.lum.ai/skema/img2mml/models"
+    REPO_NAME = "lum-ai/img2mml"
     MODEL_NAME = "cnn_xfmer_arxiv_im2mml_with_fonts_boldface_best.pt"
     # If the model path is none or doesn't exist, the default model will be downloaded from server.
     if model_path is None or not os.path.exists(model_path):
@@ -34,10 +35,9 @@ def retrieve_model(model_path=None) -> str:
         # Check if the model file already exists
         if not os.path.exists(model_path):
             # If the file doesn't exist, download it from the specified URL
-            url = f"{MODEL_BASE_ADDRESS}/{MODEL_NAME}"
-            print(f"Downloading the model checkpoint from {url}...")
-            urllib.request.urlretrieve(url, model_path)
-
+            print(f"Downloading the model checkpoint from HuggingFace...")
+            hf_hub_download(repo_id=REPO_NAME, filename=MODEL_NAME, local_dir=model_path.parent, local_dir_use_symlinks=False)
+
     return str(model_path)
 
 

skema/program_analysis/CAST/fortran/tests/test_compound_conditional_cast_fortran.py

@@ -0,0 +1,101 @@
+import pytest
+from tempfile import TemporaryDirectory
+from pathlib import Path 
+
+from skema.program_analysis.CAST.fortran.ts2cast import TS2CAST
+from skema.program_analysis.CAST2FN.model.cast import (
+    Assignment,
+    Var,
+    Name,
+    CASTLiteralValue,
+    ModelIf,
+    Operator,
+    ScalarType
+)
+
+def cond_compound1():
+    return """
+program cond_compound1
+integer :: a = 3
+if (a .gt. 1 .and. a .lt. 10) then
+    a = 40
+end if
+end program cond_compound1
+    """
+
+
+def generate_cast(test_file_string):
+    with TemporaryDirectory() as temp:
+        source_path = Path(temp) / "source.f95"
+        source_path.write_text(test_file_string)
+        out_cast = TS2CAST(str(source_path)).out_cast
+
+    return out_cast[0]
+
+def test_cond_compound1():
+    exp_cast = generate_cast(cond_compound1())
+
+    asg_node = exp_cast.nodes[0].body[0]
+
+    assert isinstance(asg_node, Assignment)
+    assert isinstance(asg_node.left, Var)
+    assert isinstance(asg_node.left.val, Name)
+    assert asg_node.left.val.name == "a"
+    assert asg_node.left.val.id == 0
+
+    assert isinstance(asg_node.right, CASTLiteralValue)
+    assert asg_node.right.value_type == ScalarType.INTEGER
+    assert asg_node.right.value == '3'
+
+    cond_node = exp_cast.nodes[0].body[1]
+    cond_expr = cond_node.expr
+    assert isinstance(cond_node, ModelIf)
+    assert isinstance(cond_expr, ModelIf)
+
+    if_node = cond_expr
+    assert isinstance(if_node, ModelIf)
+
+    expr = if_node.expr
+    assert isinstance(expr, Operator)
+    assert expr.op == ".gt."
+    assert len(expr.operands) == 2
+    assert isinstance(expr.operands[1], CASTLiteralValue)
+    assert expr.operands[1].value_type == ScalarType.INTEGER
+    assert expr.operands[1].value == "1"
+
+    assert isinstance(expr.operands[0], Name)
+    assert expr.operands[0].name == "a"
+    assert expr.operands[0].id == 0
+
+    assert len(if_node.body) == 1
+    body = if_node.body[0]
+    assert isinstance(body, Operator)
+    assert body.op == ".lt."
+    assert len(body.operands) == 2
+    assert isinstance(body.operands[0], Name)
+    assert body.operands[0].name == "a"
+    assert body.operands[0].id == 0
+
+    assert isinstance(body.operands[1], CASTLiteralValue)
+    assert body.operands[1].value_type == ScalarType.INTEGER
+    assert body.operands[1].value == "10"
+
+    assert len(if_node.orelse) == 1
+    orelse = if_node.orelse[0]
+    assert isinstance(orelse, CASTLiteralValue)
+    assert orelse.value_type == ScalarType.BOOLEAN
+    assert orelse.value == False
+
+    cond_body = cond_node.body
+    assert len(cond_body) == 1
+    assert isinstance(cond_body[0], Assignment)
+    assert isinstance(cond_body[0].left, Var)
+    assert cond_body[0].left.val.name == "a"
+    assert cond_body[0].left.val.id == 0
+
+    assert isinstance(cond_body[0].right, CASTLiteralValue)
+    assert cond_body[0].right.value_type == ScalarType.INTEGER
+    assert cond_body[0].right.value == '40'
+
+    cond_else = cond_node.orelse
+    assert len(cond_else) == 0