add gensen aml notebook testing

scenarios/sentence_similarity/gensen_aml_deep_dive.ipynb

@@ -82,6 +82,8 @@
     "import os\n",
     "import pandas as pd\n",
     "import shutil\n",
+    "import papermill as pm\n",
+    "import scrapbook as sb\n",
     "\n",
     "sys.path.append(\"../../\")\n",
     "from utils_nlp.dataset import snli, preprocess, Split\n",
@@ -116,12 +118,26 @@
   {
    "cell_type": "code",
    "execution_count": 2,
-   "metadata": {},
+   "metadata": {
+    "tags": [
+     "parameters"
+    ]
+   },
    "outputs": [],
    "source": [
+    "# Model configuration\n",
+    "NROWS = None\n",
     "CACHE_DIR = \"./temp\"\n",
     "AZUREML_CONFIG_PATH = \"./.azureml\"\n",
-    "AZUREML_VERBOSE = False  # Prints verbose azureml logs when True"
+    "AZUREML_VERBOSE = False  # Prints verbose azureml logs when True\n",
+    "MAX_EPOCH = None\n",
+    "ENTRY_SCRIPT = \"utils_nlp/gensen/gensen_train.py\"\n",
+    "TRAIN_SCRIPT = \"gensen_train.py\"\n",
+    "CONFIG_PATH = \"gensen_config.json\"\n",
+    "EXPERIMENT_NAME = \"NLP-SS-GenSen-deepdive\"\n",
+    "UTIL_NLP_PATH = \"../../utils_nlp\"\n",
+    "MAX_TOTAL_RUNS = 8\n",
+    "MAX_CONCURRENT_RUNS = 4"
    ]
   },
   {
@@ -213,9 +229,9 @@
    "outputs": [],
    "source": [
     "data_dir = os.path.join(CACHE_DIR, \"data\")\n",
-    "train = snli.load_pandas_df(data_dir, file_split=Split.TRAIN)\n",
-    "dev = snli.load_pandas_df(data_dir, file_split=Split.DEV)\n",
-    "test = snli.load_pandas_df(data_dir, file_split=Split.TEST)"
+    "train = snli.load_pandas_df(data_dir, file_split=Split.TRAIN, nrows=NROWS)\n",
+    "dev = snli.load_pandas_df(data_dir, file_split=Split.DEV, nrows=NROWS)\n",
+    "test = snli.load_pandas_df(data_dir, file_split=Split.TEST, nrows=NROWS)"
    ]
   },
   {
@@ -749,7 +765,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "_ = shutil.copytree(\"../../utils_nlp\", utils_folder)"
+    "_ = shutil.copytree(UTIL_NLP_PATH, utils_folder)"
    ]
   },
   {
@@ -758,8 +774,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "_ = shutil.copy(\"gensen_train.py\", os.path.join(utils_folder, \"gensen\"))\n",
-    "_ = shutil.copy(\"gensen_config.json\", os.path.join(utils_folder, \"gensen\"))"
+    "_ = shutil.copy(TRAIN_SCRIPT, os.path.join(utils_folder, \"gensen\"))\n",
+    "_ = shutil.copy(CONFIG_PATH, os.path.join(utils_folder, \"gensen\"))"
    ]
   },
   {
@@ -790,13 +806,14 @@
     "script_params = {\n",
     "    \"--config\": \"utils_nlp/gensen/gensen_config.json\",\n",
     "    \"--data_folder\": ws.get_default_datastore().as_mount(),\n",
+    "    \"--max_epoch\": MAX_EPOCH,\n",
     "}\n",
     "\n",
     "estimator = PyTorch(\n",
     "    source_directory=project_folder,\n",
     "    script_params=script_params,\n",
     "    compute_target=compute_target,\n",
-    "    entry_script=\"utils_nlp/gensen/gensen_train.py\",\n",
+    "    entry_script=ENTRY_SCRIPT,\n",
     "    node_count=2,\n",
     "    process_count_per_node=1,\n",
     "    distributed_training=MpiConfiguration(),\n",
@@ -842,7 +859,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "experiment_name = \"pytorch-gensen\"\n",
+    "experiment_name = EXPERIMENT_NAME\n",
     "experiment = Experiment(ws, name=experiment_name)"
    ]
   },
@@ -2029,7 +2046,7 @@
     }
    ],
    "source": [
-    "_ = run.wait_for_completion(show_output=True) # Block until the script has completed training."
+    "_ = run.wait_for_completion(show_output=AZUREML_VERBOSE) # Block until the script has completed training."
    ]
   },
   {
@@ -2084,8 +2101,8 @@
     "    policy=early_termination_policy,\n",
     "    primary_metric_name=\"min_val_loss\",\n",
     "    primary_metric_goal=PrimaryMetricGoal.MINIMIZE,\n",
-    "    max_total_runs=8,\n",
-    "    max_concurrent_runs=4,\n",
+    "    max_total_runs=MAX_TOTAL_RUNS,\n",
+    "    max_concurrent_runs=MAX_CONCURRENT_RUNS,\n",
     ")"
    ]
   },
@@ -2119,7 +2136,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "#RunDetails(hyperdrive_run).show()"
+    "RunDetails(hyperdrive_run).show()"
    ]
   },
   {
@@ -2168,13 +2185,22 @@
    "source": [
     "best_run = hyperdrive_run.get_best_run_by_primary_metric()\n",
     "best_run_metrics = best_run.get_metrics()\n",
-    "if AZUREML_VERBOSE:\n",
-    "    print(best_run)\n",
-    "    print(\n",
-    "        \"Best Run:\\n  Validation loss: {0:.5f} \\n  Learning rate: {1:.5f} \\n\".format(\n",
-    "            best_run_metrics[\"best_val_loss\"][-1], best_run_metrics[\"lr\"]\n",
-    "        )\n",
-    "    )"
+    "print(\n",
+    "    \"Best Run:\\n  Validation loss: {0:.5f} \\n  Learning rate: {1:.5f} \\n\".format(\n",
+    "        best_run_metrics[\"min_val_loss\"], best_run_metrics[\"learning_rate\"]\n",
+    "    )\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Persist properties of the run so we can access the logged metrics later\n",
+    "sb.glue(\"min_val_loss\", best_run_metrics['min_val_loss'])\n",
+    "sb.glue(\"learning_rate\", best_run_metrics['learning_rate'])"
    ]
   },
   {
@@ -2197,10 +2223,11 @@
     "name": "minxia"
    }
   ],
+  "celltoolbar": "Tags",
   "kernelspec": {
-   "display_name": "Python (nlp_cpu)",
+   "display_name": "Python 3",
    "language": "python",
-   "name": "nlp_cpu"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

scenarios/sentence_similarity/gensen_train.py

@@ -595,6 +595,7 @@ def train(config, data_folder, learning_rate=0.0001, max_epoch=None):
                     )
                     if training_complete:
                         mlflow.log_metric("min_val_loss", float(min_val_loss))
+                        mlflow.log_metric("learning_rate", learning_rate)
                         break
 
                     logging.info("Evaluating on NLI")
@@ -633,10 +634,18 @@ if __name__ == "__main__":
         help="Limit training to specified number of epochs.",
     )
 
+    parser.add_argument(
+        "--max_epoch",
+        type=int,
+        default=None,
+        help="Limit training to specified number of epochs.",
+    )
+
     args = parser.parse_args()
     data_path = args.data_folder
     lr = args.learning_rate
 
     config_file_path = args.config
+    max_epoch = args.max_epoch
     config_obj = read_config(config_file_path)
-    train(config_obj, data_path, lr)
+    train(config_obj, data_path, lr, max_epoch)

tests/conftest.py

@@ -50,6 +50,9 @@ def notebooks():
         "gensen_local": os.path.join(
             folder_notebooks, "sentence_similarity", "gensen_local.ipynb"
         ),
+        "gensen_azureml": os.path.join(
+            folder_notebooks, "sentence_similarity", "gensen_aml_deep_dive.ipynb"
+        ),
     }
     return paths