create action

Author: ninthreezy

datatorch/cli/action/create.py

@@ -3,6 +3,19 @@
 import yaml
 import json as jsonlib
 
+# What a local action's run.py starts as: the image comes in, a result
+# goes out, and the model goes in the middle.
+MODEL_SCRIPT = """import datatorch
+
+image_path = datatorch.get_input("imagePath")
+
+# Load your model and run it on image_path here. Anything your model
+# needs must be installed in the Python this runner runs with.
+result = "ok"
+
+datatorch.set_output("result", result)
+"""
+
 
 @click.command(
     help="Creates a basic hello world action in the current folder, or, with "
@@ -62,22 +75,47 @@ def create(name, description, json, local_name, tag):
         os.makedirs(target, exist_ok=True)
         if name == "Hello World":
             name = local_name
-    config = {
-        "$schema": "http://datatorch.io/schema/action.v1.json",
-        "name": name,
-        "description": description,
-        "inputs": {
-            "example": {
-                "type": "string",
-                "description": "Input string that will get printed to console.",
-                "default": "Default Example",
-            }
-        },
-        "outputs": {},
-        # Inputs reach shell commands as $INPUT_<NAME> environment variables;
-        # interpolating ${{ input.* }} into a command raises InputInjectionError.
-        "runs": {"using": "cmd", "command": 'echo "$INPUT_EXAMPLE"'},
-    }
+    if local_name is not None:
+        # A local action is almost always someone's own model: start it
+        # as one. The runner hands over the image's path; run.py returns
+        # a result. Mirrors the in-browser editor's scaffold.
+        config = {
+            "$schema": "http://datatorch.io/schema/action.v1.json",
+            "name": name,
+            "description": (
+                description
+                if description != "Action created using CI template."
+                else "Runs my model on one image."
+            ),
+            "inputs": {
+                "imagePath": {
+                    "type": "string",
+                    "required": True,
+                    "description": "Path of the image on disk",
+                }
+            },
+            "outputs": {"result": {"type": "string"}},
+            "runs": {"using": "python", "main": "run.py"},
+        }
+        with open(os.path.join(target, "run.py"), "w") as script:
+            script.write(MODEL_SCRIPT)
+    else:
+        config = {
+            "$schema": "http://datatorch.io/schema/action.v1.json",
+            "name": name,
+            "description": description,
+            "inputs": {
+                "example": {
+                    "type": "string",
+                    "description": "Input string that will get printed to console.",
+                    "default": "Default Example",
+                }
+            },
+            "outputs": {},
+            # Inputs reach shell commands as $INPUT_<NAME> environment variables;
+            # interpolating ${{ input.* }} into a command raises InputInjectionError.
+            "runs": {"using": "cmd", "command": 'echo "$INPUT_EXAMPLE"'},
+        }
     if json:
         with open(os.path.join(target, "action-datatorch.json"), "w") as outfile:
             jsonlib.dump(config, outfile, indent=2, sort_keys=True)

setup.py

@@ -33,7 +33,7 @@
 
 setup(
     name="datatorch",
-    version="0.7.5",
+    version="0.7.7",
     description="A CLI and library for interacting with DataTorch.",
     author="DataTorch",
     author_email="support@datatorch.io",