wip

Author: EC2 Default User

sagemaker/part1_model_research.ipynb

@@ -59,7 +59,7 @@
     "from glob import glob\n",
     "import boto3\n",
     "\n",
-    "# download files from s3 \"directory\"\n",
+    "# download train files from s3 \"directory\"\n",
     "s3_bucket = 'jakechenawspublic'\n",
     "s3_prefix = 'sample_data/mnist/train'\n",
     "\n",
@@ -70,10 +70,10 @@
     "s3_keys = [r['Key'] for r in response['Contents']]\n",
     "for key_name in s3_keys:\n",
     "    fname = key_name.replace(s3_prefix, '').lstrip('/') # create local file name by removing prefix from key name\n",
-    "    s3.download_file(bucket_name, key_name, fname)      # download file into training dir\n",
+    "    s3.download_file(bucket_name, key_name, fname)      # download train file\n",
     "\n",
-    "# load downloaded files into np.array\n",
-    "fnames = glob('*.csv')\n",
+    "# load downloaded files (in this case, file) into np.array\n",
+    "fnames = glob('*train.csv')\n",
     "arrays = np.array([np.loadtxt(f, delimiter=',') for f in fnames])\n",
     "\n",
     "# join files into one array with shape [records, 785]\n",
@@ -170,7 +170,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "#### Test Network"
+    "### Phase 5: Model Evaluation"
    ]
   },
   {
@@ -187,13 +187,25 @@
     }
    ],
    "source": [
-    "bucket_name = 'jakechenawspublic'\n",
-    "key_name = 'sample_data/mnist/test/mnist_test.csv'\n",
+    "batch_size = 5\n",
+    "\n",
+    "# download test files from s3 \"directory\"\n",
+    "s3_bucket = 'jakechenawspublic'\n",
+    "s3_prefix = 'sample_data/mnist/test'\n",
     "\n",
     "s3 = boto3.client('s3')\n",
-    "s3.download_file(bucket_name, key_name, 'mnist_test.csv')\n",
     "\n",
-    "mnist_test = np.loadtxt('mnist_test.csv', delimiter=',')\n",
+    "response = s3.list_objects_v2(Bucket=s3_bucket, Prefix=s3_prefix)\n",
+    "\n",
+    "s3_keys = [r['Key'] for r in response['Contents']]\n",
+    "for key_name in s3_keys:\n",
+    "    fname = key_name.replace(s3_prefix, '').lstrip('/') # create local file name by removing prefix from key name\n",
+    "    s3.download_file(bucket_name, key_name, fname)      # download test file\n",
+    "\n",
+    "fnames = glob('*_test.csv')\n",
+    "arrays = np.array([np.loadtxt(f, delimiter=',') for f in fnames])\n",
+    "\n",
+    "mnist_test = arrays.reshape(-1, 785)\n",
     "X_test = mnist_test.T[1:].T.reshape(-1,1,28,28)\n",
     "y_test = mnist_test.T[:1].T.reshape(-1)\n",
     "\n",
@@ -203,7 +215,10 @@
     "acc = mx.metric.Accuracy()\n",
     "lenet_model.score(test_iter, acc)\n",
     "print(acc)\n",
-    "assert acc.get()[1] > 0.98"
+    "assert acc.get()[1] > 0.98\n",
+    "\n",
+    "# predict function for lenet\n",
+    "lenet_model.predict(test_iter)"
    ]
   },
   {

sagemaker/part2_sm_mnist.py

@@ -122,13 +122,7 @@ def train(
     fc2 = mx.sym.FullyConnected(data=tanh3, num_hidden=10)
     # softmax loss
     lenet = mx.sym.SoftmaxOutput(data=fc2, name='softmax')
-    # define training batch size
-    batch_size = 100
 
-    # create iterator around training and validation data
-    train_iter = mx.io.NDArrayIter(mnist['train_data'][:ntrain], mnist['train_label'][:ntrain], batch_size, shuffle=True)
-    val_iter = mx.io.NDArrayIter(mnist['train_data'][ntrain:], mnist['train_label'][ntrain:], batch_size)
-    
     """
     End copy/paste from tutorial part 1
     """
@@ -153,7 +147,7 @@ def train(
                     kvstore=kvstore # added kvstore argument
                    )
 
-    return lenet
+    return lenet_model
 
 
 # ---------------------------------------------------------------------------- #