Fix caffe2 predict (facebookresearch#1103)

Summary:
Pull Request resolved: facebookresearch#1103

Currently it doesn't correctly handle str and non-str input. This commit fix the issue

Differential Revision: D18286625

fbshipit-source-id: 91a4e33d535f854c679e913bc4946098f1f7e4c5

pytext/__init__.py

@@ -32,9 +32,15 @@ def _predict(workspace_id, predict_net, model, tensorizers, input):
     }
     model_inputs = model.arrange_model_inputs(tensor_dict)
     model_input_names = model.get_export_input_names(tensorizers)
+    vocab_to_export = model.vocab_to_export(tensorizers)
     for blob_name, model_input in zip(model_input_names, model_inputs):
-        converted_blob_name = convert_caffe2_blob_name(blob_name)
-        workspace.blobs[converted_blob_name] = np.array([model_input], dtype=str)
+        converted_blob_name = blob_name
+        dtype = np.float32
+        if blob_name in vocab_to_export:
+            converted_blob_name = convert_caffe2_blob_name(blob_name)
+            dtype = str
+
+        workspace.blobs[converted_blob_name] = np.array([model_input], dtype=dtype)
     workspace.RunNet(predict_net)
     return {
         str(blob): workspace.blobs[blob][0] for blob in predict_net.external_outputs

tests/data/test_data_tiny.tsv

@@ -1,4 +1,4 @@
-alarm/set_alarm	11:17:datetime	reactivate weekly alarm	
-alarm/set_alarm	23:38:datetime	Set alarm to ring only on the weekdays	
-alarm/time_left_on_alarm		When will alarm go off	
-reminder/set_reminder	10:18:datetime,22:38:reminder/todo	remind me tomorrow to call the groomer	
+alarm/set_alarm	11:17:datetime	reactivate weekly alarm	[1.0]
+alarm/set_alarm	23:38:datetime	Set alarm to ring only on the weekdays	[1.0]
+alarm/time_left_on_alarm		When will alarm go off	[1.0]
+reminder/set_reminder	10:18:datetime,22:38:reminder/todo	remind me tomorrow to call the groomer	[1.0]

tests/data/train_data_tiny.tsv

@@ -1,10 +1,10 @@
-alarm/modify_alarm	16:24:datetime,39:57:datetime	change my alarm tomorrow to wake me up 30 minutes earlier	
-alarm/set_alarm		Turn on all my alarms	
-alarm/set_alarm	12:27:datetime	sound alarm every 8 minutes	
-alarm/set_alarm	7:17:datetime	repeat yesterdays alarm	
-alarm/snooze_alarm		continue my alarm	
-alarm/time_left_on_alarm		Do I have anymore time on the alarm	
-reminder/set_reminder	10:22:datetime,28:56:reminder/todo	remind me Monday night that get doe with work 12 Tuesday	
-reminder/show_reminders	8:15:datetime,18:27:reminder/noun	display Tuesday's reminders	
-weather/find	12:19:weather/noun,20:28:datetime	what is the weather tomorrow	
-weather/find	13:17:weather/attribute	When will it snow	
+alarm/modify_alarm	16:24:datetime,39:57:datetime	change my alarm tomorrow to wake me up 30 minutes earlier	[1.0]
+alarm/set_alarm		Turn on all my alarms	[1.0]
+alarm/set_alarm	12:27:datetime	sound alarm every 8 minutes	[1.0]
+alarm/set_alarm	7:17:datetime	repeat yesterdays alarm	[1.0]
+alarm/snooze_alarm		continue my alarm	[1.0]
+alarm/time_left_on_alarm		Do I have anymore time on the alarm	[1.0]
+reminder/set_reminder	10:22:datetime,28:56:reminder/todo	remind me Monday night that get doe with work 12 Tuesday	[1.0]
+reminder/show_reminders	8:15:datetime,18:27:reminder/noun	display Tuesday's reminders	[1.0]
+weather/find	12:19:weather/noun,20:28:datetime	what is the weather tomorrow	[1.0]
+weather/find	13:17:weather/attribute	When will it snow	[1.0]

tests/predictor_test.py

@@ -5,10 +5,17 @@
 import tempfile
 import unittest
 
+import numpy as np
 from pytext import batch_predict_caffe2_model
 from pytext.config import LATEST_VERSION, PyTextConfig
 from pytext.data import Data
 from pytext.data.sources import TSVDataSource
+from pytext.data.tensorizers import (
+    FloatListTensorizer,
+    LabelTensorizer,
+    TokenTensorizer,
+)
+from pytext.models.doc_model import DocModel
 from pytext.task import create_task
 from pytext.task.serialize import save
 from pytext.task.tasks import DocumentClassificationTask
@@ -25,14 +32,23 @@ def test_batch_predict_caffe2_model(self):
             eval_data = tests_module.test_file("test_data_tiny.tsv")
             config = PyTextConfig(
                 task=DocumentClassificationTask.Config(
+                    model=DocModel.Config(
+                        inputs=DocModel.Config.ModelInput(
+                            tokens=TokenTensorizer.Config(),
+                            dense=FloatListTensorizer.Config(
+                                column="dense", dim=1, error_check=True
+                            ),
+                            labels=LabelTensorizer.Config(),
+                        )
+                    ),
                     data=Data.Config(
                         source=TSVDataSource.Config(
                             train_filename=train_data,
                             eval_filename=eval_data,
                             test_filename=eval_data,
-                            field_names=["label", "slots", "text"],
+                            field_names=["label", "slots", "text", "dense"],
                         )
-                    )
+                    ),
                 ),
                 version=LATEST_VERSION,
                 save_snapshot_path=snapshot_file.name,
@@ -47,3 +63,12 @@ def test_batch_predict_caffe2_model(self):
                 snapshot_file.name, caffe2_model_file.name
             )
             self.assertEqual(4, len(results))
+
+            pt_results = task.predict(task.data.data_source.test)
+
+            for pt_res, res in zip(pt_results, results):
+                print(pt_res["score"])
+                print(res)
+                np.testing.assert_array_almost_equal(
+                    pt_res["score"].tolist()[0], [score[0] for score in res.values()]
+                )