remove DocModel_Deprecated (#916)

Summary:
Pull Request resolved: #916

Removing DocClassificationTask_Deprecated and DocModel_Deprecated.  They are
replaced with DocumentClassificationTask and DocModel, respectively,
introduced in v0.2.0

Reviewed By: borguz

Differential Revision: D16847219

fbshipit-source-id: 8254806ec708d773acf1aceda3179c7a02ad5b7f

pytext/builtin_task.py

@@ -13,7 +13,6 @@
 from pytext.task.task import Task_Deprecated
 from pytext.task.tasks import (
     BertPairRegressionTask,
-    DocClassificationTask_Deprecated,
     DocumentClassificationTask,
     DocumentRegressionTask,
     EnsembleTask,
@@ -66,7 +65,6 @@ def register_builtin_tasks():
         (
             BertPairRegressionTask,
             DisjointMultitask,
-            DocClassificationTask_Deprecated,
             DocumentClassificationTask,
             DocumentRegressionTask,
             EnsembleTask,

pytext/config/config_adapter.py

@@ -598,10 +598,13 @@ def migrate_to_new_data_handler(task, columns):
         lambda x: [{"filepath": x}],
     )
 
+    rename_parameter(task, "labels.label_weights", "model.output_layer.label_weights")
+
     delete_parameter(task, "data_handler")
     delete_parameter(task, "exporter")
     delete_parameter(task, "features")
     delete_parameter(task, "featurizer")
+    delete_parameter(task, "labels")
 
 
 @register_adapter(from_version=15)
@@ -614,6 +617,31 @@ def remove_lmtask_deprecated(json_config):
     return json_config
 
 
+@register_adapter(from_version=16)
+def remove_docclassificationtask_deprecated(json_config):
+    for section in find_dicts_containing_key(
+        json_config, "DocClassificationTask_Deprecated"
+    ):
+        task = section.pop("DocClassificationTask_Deprecated")
+        convert = next(find_dicts_containing_key(task, "convert_to_bytes"), None)
+
+        section["DocumentClassificationTask"] = task
+        migrate_to_new_data_handler(task, ["doc_label", "text"])
+        create_parameter(task, "model.inputs.labels.column", "doc_label")
+
+        # In DocumentClassificationTask.Config:
+        #   model: BaseModel.Config = DocModel.Config()
+        # It will create a BaseModel if model class is implicit in json.
+        # We make it explicit to avoid errors.
+        for model in find_dicts_containing_key(section, "model"):
+            if next(iter(model["model"]))[0].islower():
+                model["model"] = {"DocModel": model.pop("model")}
+
+        if convert and convert["convert_to_bytes"]:
+            rename(section, "DocModel", "ByteTokensDocumentModel")
+    return json_config
+
+
 def upgrade_one_version(json_config):
     current_version = json_config.get("version", 0)
     adapter = ADAPTERS.get(current_version)

pytext/config/pytext_config.py

@@ -143,4 +143,4 @@ class TestConfig(ConfigBase):
     test_out_path: str = ""
 
 
-LATEST_VERSION = 16
+LATEST_VERSION = 17

pytext/config/test/json_config/v16.json

@@ -0,0 +1,154 @@
+[
+  {
+    "original": {
+      "task": {
+        "DocClassificationTask_Deprecated": {
+          "data_handler": {
+            "train_path": "tests/data/train_data_tiny.tsv",
+            "eval_path": "tests/data/test_data_tiny.tsv",
+            "test_path": "tests/data/test_data_tiny.tsv"
+          },
+          "model": {
+            "representation": {
+              "DocNNRepresentation": {}
+            }
+          }
+        }
+      },
+      "version": 16
+    },
+    "adapted": {
+      "task": {
+        "DocumentClassificationTask": {
+          "data": {
+            "source": {
+              "TSVDataSource": {
+                "eval_filename": "tests/data/test_data_tiny.tsv",
+                "test_filename": "tests/data/test_data_tiny.tsv",
+                "train_filename": "tests/data/train_data_tiny.tsv",
+                "field_names": ["doc_label", "text"]
+              }
+            },
+            "batcher": {
+              "PoolingBatcher": {}
+            }
+          },
+          "model": {
+            "DocModel": {
+              "inputs": {
+                "labels": {
+                  "column": "doc_label"
+                }
+              },
+              "representation": {
+                "DocNNRepresentation": {}
+              }
+            }
+          }
+        }
+      },
+      "version": 17
+    }
+  },
+  {
+    "original": {
+      "task": {
+        "DocClassificationTask_Deprecated": {
+          "data_handler": {
+            "columns_to_read": ["label", "raw_text"],
+            "train_path": "tests/data/train_data_tiny.tsv",
+            "eval_path": "tests/data/test_data_tiny.tsv",
+            "test_path": "tests/data/test_data_tiny.tsv"
+          }
+        }
+      },
+      "version": 16
+    },
+    "adapted": {
+      "task": {
+        "DocumentClassificationTask": {
+          "data": {
+            "source": {
+              "TSVDataSource": {
+                "eval_filename": "tests/data/test_data_tiny.tsv",
+                "test_filename": "tests/data/test_data_tiny.tsv",
+                "train_filename": "tests/data/train_data_tiny.tsv",
+                "field_names": ["label", "raw_text"]
+              }
+            },
+            "batcher": {
+              "PoolingBatcher": {}
+            }
+          },
+          "model": {
+            "DocModel": {
+              "inputs": {
+                "labels": {
+                  "column": "doc_label"
+                }
+              }
+            }
+          }
+        }
+      },
+      "version": 17
+    }
+  },
+  {
+    "original": {
+      "task": {
+        "DocClassificationTask_Deprecated": {
+          "data_handler": {
+            "train_path": "tests/data/train_data_tiny.tsv",
+            "eval_path": "tests/data/test_data_tiny.tsv",
+            "test_path": "tests/data/test_data_tiny.tsv"
+          },
+          "featurizer": {
+            "SimpleFeaturizer": {
+              "convert_to_bytes": true,
+              "lowercase_tokens": false
+            }
+          },
+          "model": {
+            "representation": {
+              "DocNNRepresentation": {}
+            }
+          }
+        }
+      },
+      "version": 16
+    },
+    "adapted": {
+      "task": {
+        "DocumentClassificationTask": {
+          "data": {
+            "source": {
+              "TSVDataSource": {
+                "eval_filename": "tests/data/test_data_tiny.tsv",
+                "test_filename": "tests/data/test_data_tiny.tsv",
+                "train_filename": "tests/data/train_data_tiny.tsv",
+                "field_names": ["doc_label", "text"]
+              }
+            },
+            "batcher": {
+              "PoolingBatcher": {}
+            }
+          },
+          "model": {
+            "ByteTokensDocumentModel": {
+              "inputs": {
+                "labels": {
+                  "column": "doc_label"
+                }
+              },
+              "representation": {
+                "DocNNRepresentation": {}
+              }
+            }
+          }
+        }
+      },
+      "version": 17
+    }
+  }
+]

pytext/data/__init__.py

@@ -12,7 +12,6 @@
 from .data_handler import BatchIterator, CommonMetadata, DataHandler
 from .disjoint_multitask_data import DisjointMultitaskData
 from .disjoint_multitask_data_handler import DisjointMultitaskDataHandler
-from .doc_classification_data_handler import DocClassificationDataHandler, RawData
 from .tensorizers import Tensorizer
 
 
@@ -26,12 +25,10 @@
     "DataHandler",
     "DisjointMultitaskData",
     "DisjointMultitaskDataHandler",
-    "DocClassificationDataHandler",
     "EvalBatchSampler",
     "generator_iterator",
     "PoolingBatcher",
     "RandomizedBatchSampler",
-    "RawData",
     "RoundRobinBatchSampler",
     "Tensorizer",
 ]