pushing and eating

notebooks/FinalMilestone.ipynb

@@ -118673,7 +118673,7 @@
    "id": "fb769a60",
    "metadata": {},
    "source": [
-    "Now let's do some predictions on the test set for the hot data.  The mean absolute percentage error is 0.471, which we will explain below after we predict on the cold data."
+    "Now let's do some predictions on the test set for the hot data.  The mean absolute percentage error is 0.471, which we will explain below after we model the cold data."
    ]
   },
   {
@@ -118773,7 +118773,7 @@
     "\n",
     "cold_X_train, cold_X_test, cold_y_train, cold_y_test = train_test_split(cold_X, cold_y, test_size = 0.20, random_state = 0)\n",
     "\n",
-    "coldgam = PoissonGAM()\n",
+    "coldgam = PoissonGAM() \n",
     "\n",
     "lam_grid = [0, 0.0001, 0.0002, 0.0003, 0.0004, 0.0005, 0.0006, 0.0007, 0.0008, 0.0009, 0.001]\n",
     "splines_grid = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]\n",
@@ -118783,6 +118783,14 @@
     "print(coldgam.summary())"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "97bbd835",
+   "metadata": {},
+   "source": [
+    "We then calculate and print the MAPE of the cold GAM model below."
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 236,
@@ -118813,6 +118821,14 @@
     "print(mape)\n"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "c4537ded",
+   "metadata": {},
+   "source": [
+    "Now we compare each model's ability to predict the violent crime counts on the entire dataset. We use the mean absolute percentage error because it better explains the difference in forecasting between the two models. As you see below, the hot model has a much lower MAPE than the cold model despite them having similar mean absolute errors. This suggest their forecasting abilities differ drastically, which suggest less of a predictive utility for this model.  "
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 237,
@@ -120475,14 +120491,16 @@
    "id": "e6d691c5-349f-4a11-b172-f27753329f59",
    "metadata": {},
    "source": [
-    "Our Generalized Addititive Model does an effective job of fitting to the data, as demonstrated above by the psuedo R-square parameter. Perhaps this is due to our high effective degrees of freedom, which gets into the trade-off of bias versus variance, but we attempted to minimize this as much as we could. Poisson GAM was our most effective model due to the Poissonian error distribution conforming nicely to the errors associated with a strictly-positive count variable. We are able to explain much of the variance when it comes to "
+    "Our Generalized Addititive Models do an effective job of fitting to the data, as demonstrated above by the psuedo R-square parameter. Perhaps this is due to our high effective degrees of freedom, which gets into the trade-off of bias versus variance, but we attempted to minimize that as much as we could. Poisson GAM was our most effective model due to the Poissonian error distribution conforming nicely to the errors associated with a strictly-positive count variable. We are able to explain much of the variance when it comes to the violent crime counts per day based on the temperature, but the predictive utility is obviously not very good. This is expected, as we were primarily looking for a relationship between weather factors and crime, but not intending to do crime forecasting. That is an entire area of research that would require much more data than the weather on a given day. "
    ]
   },
   {
    "cell_type": "markdown",
    "id": "48c9cc01",
    "metadata": {},
-   "source": []
+   "source": [
+    "Not finished yet..."
+   ]
   }
  ],
  "metadata": {