Merge pull request #739 from erikcs/maq-optional

Remove maq git+pip dependency

causalml/metrics/__init__.py

@@ -32,4 +32,3 @@
     SensitivitySubsetData,
     SensitivitySelectionBias,
 )  # noqa
-from maq import MAQ, get_ipw_scores  # noqa

docs/examples/qini_curves_for_costly_treatment_arms.ipynb

@@ -7,7 +7,25 @@
    "source": [
     "# Qini curves with multiple costly treatment arms\n",
     "\n",
-    "This notebook gives a brief overview of Qini curves for multi-armed treatment rules and a simple simulated example."
+    "This notebook shows approaches to evaluating multi-armed CATE estimators from `causalML` with the Multi-Armed Qini metric available in the `maq` package (available at https://github.com/grf-labs/maq).\n",
+    "\n",
+    "\n",
+    "This metric is a generalization of the familiar *Qini curve* to settings where we have multiple treatment arms available, and the cost of assigning treatment can vary by both unit and treatment arm according to some known cost structure. At a high level, this metric essentially allows you to quantify the value of targeting with more treatment arms by undertaking a cost-benefit exercise that uses your CATE estimates to assign the arm to the unit that is most cost-beneficial at various budget constraints.\n",
+    "\n",
+    "This notebook gives a brief overview of the statistical setup and a walkthrough with a simple simulated example. \n",
+    "\n",
+    "\n",
+    "To use this functionality, you first have to install the `maq` Python package from GitHub. The latest source release can be installed with:"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "0a633fa7",
+   "metadata": {},
+   "source": [
+    "```\n",
+    "pip install \"git+https://github.com/grf-labs/maq.git#egg=maq&subdirectory=python-package\"\n",
+    "```"
    ]
   },
   {
@@ -22,7 +40,7 @@
     "from xgboost import XGBRFRegressor\n",
     "\n",
     "# Generalized Qini curves\n",
-    "from causalml.metrics import MAQ, get_ipw_scores\n",
+    "from maq import MAQ, get_ipw_scores\n",
     "\n",
     "import numpy as np\n",
     "np.random.seed(42)"

pyproject.toml

@@ -45,7 +45,6 @@ dependencies = [
     "torch",
     "pyro-ppl",
     "graphviz",
-    "maq@git+https://github.com/grf-labs/maq.git@py0.2.2#egg=maq&subdirectory=python-package",
 ]
 
 [project.optional-dependencies]

tests/test_metrics.py

@@ -1,8 +1,6 @@
 import pandas as pd
-import numpy as np
 from numpy import isclose
 from causalml.metrics.visualize import qini_score
-from causalml.metrics import MAQ, get_ipw_scores
 
 
 def test_qini_score():
@@ -28,18 +26,3 @@ def test_qini_score():
     # for each learner, its qini score should stay same no matter calling with another model or calling separately
     assert isclose(full_result["learner_1"], learner_1_result["learner_1"])
     assert isclose(full_result["learner_2"], learner_2_result["learner_2"])
-
-
-def test_MAQ():
-    np.random.seed(42)
-    n = 1000
-    K = 5
-    tau_hat = np.random.randn(n, K)
-    cost = np.random.rand(n, K)
-    DR_scores = np.random.randn(n, K)
-
-    mq = MAQ(n_bootstrap=200)
-    mq.fit(tau_hat, cost, DR_scores)
-
-    #    (0.005729002695991717, 0.019814651108894354)
-    assert isclose(mq.average_gain(spend=0.1)[0], 0.005729)