Skip to content

Commit

Permalink
updates to model card example to include preprocessing function example
Browse files Browse the repository at this point in the history
  • Loading branch information
djm21 committed Oct 28, 2024
1 parent d55f57d commit 2bd7d80
Showing 1 changed file with 61 additions and 218 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -493,7 +493,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"The data is looking better, but the education statuses are a bit too granular. Lets combine some of them to make the job easier for our model."
"The data is looking better, but the martial status, education and work class statuses are a bit too granular. Lets combine some of them to make the job easier for our model."
]
},
{
Expand Down Expand Up @@ -549,6 +549,53 @@
"df.columns"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"If we want the automatically generated score code to leaverage these steps when scoring new data, we can put them in a preprocessing function and pass them into our import_model function call. "
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"def preprocess_function(df):\n",
" cat_vals = df[[\"WorkClass\", \"Education\", \"MartialStatus\", \"Relationship\", \"Race\", \"Sex\"]]\n",
" df = pd.get_dummies(df, columns=[\"WorkClass\", \"Education\", \"MartialStatus\", \"Relationship\", \"Race\", \"Sex\"])\n",
" df.columns = df.columns.str.replace(' ', '')\n",
" df.columns = df.columns.str.replace('-', '_')\n",
" df = df.drop(['Sex_Male'], axis=1)\n",
" df = pd.concat([df, cat_vals], axis=1).drop('index', axis=1)\n",
" # For the model to score correctly, all OHE columns must exist\n",
" input_cols = [\n",
" \"Education_9th\", \"Education_10th\", \"Education_11th\", \"Education_12th\", \"Education_Assoc_voc\", \"Education_Assoc_acdm\", \"Education_Masters\", \"Education_Prof_school\",\n",
" \"Education_Doctorate\", \"Education_Preschool\", \"Education_1st_4th\", \"Education_5th_6th\", \"Education_7th_8th\", \"WorkClass_Self_emp_inc\", \"WorkClass_Self_emp_not_inc\",\n",
" \"WorkClass_Federal_gov\", \"WorkClass_Local_gov\", \"WorkClass_State_gov\", \"WorkClass_Without_pay\", \"WorkClass_Never_worked\", \"MartialStatus_Married_spouse_absent\",\n",
" \"MartialStatus_Married_AF_spouse\", 'MartialStatus_Married_civ_spouse', 'MartialStatus_Never_married', 'MartialStatus_Divorced', 'MartialStatus_Separated', \n",
" 'MartialStatus_Widowed', 'Race_White', 'Race_Black', 'Race_Asian_Pac_Islander', 'Race_Amer_Indian_Eskimo', 'Race_Other', 'Relationship_Husband', \n",
" 'Relationship_Not_in_family', 'Relationship_Own_child', 'Relationship_Unmarried', 'Relationship_Wife', 'Relationship_Other_relative', 'WorkClass_Private',\n",
" 'Education_Bachelors'\n",
" ]\n",
" for col in input_cols:\n",
" if col not in df.columns:\n",
" df[col] = 0\n",
" df[\"Education_Some_HS\"] = df[\"Education_9th\"] | df[\"Education_10th\"] | df[\"Education_11th\"] | df[\"Education_12th\"]\n",
" df[\"Education_Assoc\"] = df[\"Education_Assoc_voc\"] | df[\"Education_Assoc_acdm\"]\n",
" df[\"Education_Adv_Degree\"] = df[\"Education_Masters\"] | df[\"Education_Prof_school\"] | df[\"Education_Doctorate\"]\n",
" df[\"Education_No_HS\"] = df[\"Education_Preschool\"] | df[\"Education_1st_4th\"] | df[\"Education_5th_6th\"] | df[\"Education_7th_8th\"]\n",
"\n",
" df[\"WorkClass_Self\"] = df[\"WorkClass_Self_emp_inc\"] | df[\"WorkClass_Self_emp_not_inc\"]\n",
" df[\"WorkClass_Gov\"] = df[\"WorkClass_Federal_gov\"] | df[\"WorkClass_Local_gov\"] | df[\"WorkClass_State_gov\"]\n",
" df[\"WorkClass_Other\"] = df[\"WorkClass_Without_pay\"] | df[\"WorkClass_Never_worked\"]\n",
"\n",
" df[\"MartialStatus_Other\"] = df[\"MartialStatus_Married_spouse_absent\"] | df[\"MartialStatus_Married_AF_spouse\"]\n",
"\n",
" return df"
]
},
{
"cell_type": "markdown",
"metadata": {},
Expand Down Expand Up @@ -663,9 +710,8 @@
"1. Write model performance statistics to a file. \n",
"1. Assess model bias and fairness (if a potentially sensitive variable is available for assessment). \n",
"1. Generate the last few model card files. \n",
"1. Write model score code (Writing the score code yourself is not necessary if you are using the import_model function to automatically generate the score code. Since we've heavily processed the data, I want to write this score code manually to include the data preprocessing code. ). \n",
"1. Generate requirements file. \n",
"1. Import model to SAS Model Manager. \n",
"1. Import model to SAS Model Manager and automatically generate the score code. \n",
"1. Open the model in SAS Model Manager and begin managing the model lifecycle there. \n",
"\n",
"So first, be sure that the variables in the block below match your use case."
Expand Down Expand Up @@ -1608,194 +1654,6 @@
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"I'm going to write the score code myself because I want to include all the data preprocessing we did earlier, but that is not a requirement. If you prefer that this score code be automatically written, I'll provide example code before the conclusion you can use instead. "
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [],
"source": [
"# Step 13: Write score code\n",
"sn = 'score_' + model_prefix + \".py\"\n",
"sc = Path.cwd() / output_path / sn"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"%%writefile $sc\n",
"\n",
"import math\n",
"import pickle\n",
"import pandas as pd\n",
"import numpy as np\n",
"from pathlib import Path\n",
"\n",
"import settings\n",
"\n",
"with open(Path(settings.pickle_path) / \"dtc.pickle\", \"rb\") as pickle_model:\n",
" model = pickle.load(pickle_model)\n",
"\n",
"def score(Age, WorkClass, Education, MartialStatus, Relationship, Race, Sex, HoursPerWeek): \n",
" \"Output: EM_EVENTPROBABILITY, EM_CLASSIFICATION\" \n",
"\n",
" # Check if pickle file is loaded, load if unavailable\n",
" try:\n",
" global model\n",
" except NameError:\n",
" with open(settings.pickle_path + 'dtc.pickle', 'rb') as _pFile:\n",
" model = pickle.load(_pFile)\n",
"\n",
" # Encode WorkClass\n",
" WorkClass_Private = 0\n",
" WorkClass_Self = 0\n",
" WorkClass_Gov = 0 \n",
" WorkClass_Other = 0\n",
" if \"Private\" in WorkClass: WorkClass_Private = 1\n",
" elif \"Self-emp-not-inc\" in WorkClass: WorkClass_Self = 1\n",
" elif \"Self-emp-inc\" in WorkClass: WorkClass_Self = 1\n",
" elif \"Local-gov\" in WorkClass: WorkClass_Gov = 1\n",
" elif \"State-gov\" in WorkClass: WorkClass_Gov = 1\n",
" elif \"Federal-gov\" in WorkClass: WorkClass_Gov = 1\n",
" else: WorkClass_Other = 1\n",
"\n",
"\n",
" # Encode Education\n",
" Education_HS_grad = 0\n",
" Education_Some_HS = 0\n",
" Education_Assoc = 0\n",
" Education_Some_college = 0\n",
" Education_Bachelors = 0\n",
" Education_Adv_Degree = 0\n",
" Education_No_HS = 0\n",
"\n",
" if \"HS-grad\" in Education: Education_HS_grad = 1\n",
" elif \"11th\" in Education: Education_Some_HS = 1\n",
" elif \"10th\" in Education: Education_Some_HS = 1\n",
" elif \"9th\" in Education: Education_Some_HS = 1\n",
" elif \"12th\" in Education: Education_Some_HS = 1\n",
" elif \"Assoc-voc\" in Education: Education_Assoc = 1\n",
" elif \"Assoc-acdm\" in Education: Education_Assoc = 1\n",
" elif \"Some-college\" in Education: Education_Some_college = 1\n",
" elif \"Bachelors\" in Education: Education_Bachelors = 1\n",
" elif \"Masters\" in Education: Education_Adv_Degree = 1\n",
" elif \"Prof-school\" in Education: Education_Adv_Degree = 1\n",
" elif \"Doctorate\" in Education: Education_Adv_Degree = 1\n",
" else:Education_No_HS = 1\n",
" \n",
" # Encode MaritalStatus\n",
" MartialStatus_Married_civ_spouse = 0\n",
" MartialStatus_Never_married = 0\n",
" MartialStatus_Divorced = 0\n",
" MartialStatus_Separated = 0 \n",
" MartialStatus_Widowed = 0 \n",
" MartialStatus_Other = 0\n",
" if \"Married-civ-spouse\" in MartialStatus: MartialStatus_Married_civ_spouse = 1\n",
" elif \"Never-married\" in MartialStatus: MartialStatus_Never_married = 1\n",
" elif \"Divorced\" in MartialStatus: MartialStatus_Divorced = 1\n",
" elif \"Separated\" in MartialStatus: MartialStatus_Separated = 1\n",
" elif \"Widowed\" in MartialStatus: MartialStatus_Widowed = 1\n",
" else: MartialStatus_Other = 1\n",
" \n",
" # Encode Relationship\n",
" Relationship_Husband = 0\n",
" Relationship_Not_in_family = 0\n",
" Relationship_Own_child= 0\n",
" Relationship_Unmarried = 0 \n",
" Relationship_Wife = 0\n",
" Relationship_Other_relative = 0\n",
" if \"Husband\" in Relationship: Relationship_Husband = 1\n",
" elif \"Not-in-family\" in Relationship: Relationship_Not_in_family = 1\n",
" elif \"Own-child\" in Relationship: Relationship_Own_child = 1\n",
" elif \"Unmarried\" in Relationship: Relationship_Unmarried = 1\n",
" elif \"Wife\" in Relationship: Relationship_Wife = 1\n",
" else: Relationship_Other_relative = 1\n",
"\n",
" # Encode Race\n",
" Race_White = 0\n",
" Race_Black = 0\n",
" Race_Asian_Pac_Islander = 0\n",
" Race_Amer_Indian_Eskimo = 0\n",
" Race_Other = 0\n",
" if \"White\" in Race: Race_White = 1\n",
" elif \"Black\" in Race: Race_Black = 1\n",
" elif \"Asian-Pac-Islander\" in Race: Race_Asian_Pac_Islander = 1\n",
" elif \"Amer-Indian-Eskimo\" in Race: Race_Amer_Indian_Eskimo = 1\n",
" else: Race_Other = 1\n",
"\n",
" # Encode Sex\n",
" Sex_Female = 0\n",
" if \"Female\" in Sex: Sex_Female = 1\n",
"\n",
" try: \n",
" input_array = pd.DataFrame([[Age, HoursPerWeek, WorkClass_Private, WorkClass_Self, WorkClass_Gov, WorkClass_Other, \n",
" Education_HS_grad, Education_Some_HS, Education_Assoc, Education_Some_college, \n",
" Education_Bachelors, Education_Adv_Degree, Education_No_HS, \n",
" MartialStatus_Married_civ_spouse, MartialStatus_Never_married, \n",
" MartialStatus_Divorced, MartialStatus_Separated, MartialStatus_Widowed, \n",
" MartialStatus_Other, Relationship_Husband, Relationship_Not_in_family, \n",
" Relationship_Own_child, Relationship_Unmarried, Relationship_Wife, \n",
" Relationship_Other_relative, Race_White, Race_Black, Race_Asian_Pac_Islander, \n",
" Race_Amer_Indian_Eskimo, Race_Other, Sex_Female]], \n",
" columns = ['Age', 'HoursPerWeek', 'WorkClass_Private', 'WorkClass_Self', 'WorkClass_Gov', 'WorkClass_Other', \n",
" 'Education_HS_grad', 'Education_Some_HS', 'Education_Assoc', 'Education_Some_college', \n",
" 'Education_Bachelors','Education_Adv_Degree', 'Education_No_HS', \n",
" 'MartialStatus_Married_civ_spouse', 'MartialStatus_Never_married', \n",
" 'MartialStatus_Divorced', 'MartialStatus_Separated', 'MartialStatus_Widowed', \n",
" 'MartialStatus_Other', 'Relationship_Husband', 'Relationship_Not_in_family', \n",
" 'Relationship_Own_child', 'Relationship_Unmarried', 'Relationship_Wife', \n",
" 'Relationship_Other_relative', 'Race_White', 'Race_Black', 'Race_Asian_Pac_Islander', \n",
" 'Race_Amer_Indian_Eskimo', 'Race_Other','Sex_Female'], \n",
" dtype = float)\n",
" except ValueError:\n",
" input_array = pd.DataFrame([[1.0, Age, HoursPerWeek, WorkClass_Private, WorkClass_Self, WorkClass_Gov, WorkClass_Other, \n",
" Education_HS_grad, Education_Some_HS, Education_Assoc, Education_Some_college, \n",
" Education_Bachelors, Education_Adv_Degree, Education_No_HS, \n",
" MartialStatus_Married_civ_spouse, MartialStatus_Never_married, \n",
" MartialStatus_Divorced, MartialStatus_Separated, MartialStatus_Widowed, \n",
" MartialStatus_Other, Relationship_Husband, Relationship_Not_in_family, \n",
" Relationship_Own_child, Relationship_Unmarried, Relationship_Wife, \n",
" Relationship_Other_relative, Race_White, Race_Black, Race_Asian_Pac_Islander, \n",
" Race_Amer_Indian_Eskimo, Race_Other, Sex_Female]], \n",
" columns = ['const', 'Age', 'HoursPerWeek', 'WorkClass_Private', 'WorkClass_Self', 'WorkClass_Gov', 'WorkClass_Other', \n",
" 'Education_HS_grad', 'Education_Some_HS', 'Education_Assoc', 'Education_Some_college', \n",
" 'Education_Bachelors', 'Education_Adv_Degree', 'Education_No_HS', \n",
" 'MartialStatus_Married_civ_spouse', 'MartialStatus_Never_married', \n",
" 'MartialStatus_Divorced', 'MartialStatus_Separated', 'MartialStatus_Widowed', \n",
" 'MartialStatus_Other', 'Relationship_Husband','Relationship_Not_in_family', \n",
" 'Relationship_Own_child', 'Relationship_Unmarried', 'Relationship_Wife', \n",
" 'Relationship_Other_relative', 'Race_White', 'Race_Black', 'Race_Asian_Pac_Islander', \n",
" 'Race_Amer_Indian_Eskimo', 'Race_Other','Sex_Female'], \n",
" dtype = float)\n",
"\n",
" prediction = model.predict_proba(input_array).tolist()\n",
"\n",
" # Check for numpy values and convert to a CAS readable representation\n",
" if isinstance(prediction, np.ndarray):\n",
" prediction = prediction.tolist()\n",
"\n",
" if input_array.shape[0] == 1:\n",
" if prediction[0][1] > 0.5:\n",
" EM_CLASSIFICATION = \"1\"\n",
" else:\n",
" EM_CLASSIFICATION = \"0\"\n",
" return EM_CLASSIFICATION, prediction[0][1]\n",
" else:\n",
" df = pd.DataFrame(prediction)\n",
" proba = df[1]\n",
" classifications = np.where(df[1] > 0.5, '1', '0')\n",
" return pd.DataFrame({'EM_CLASSIFICATION': classifications, 'EM_EVENTPROBABILITY': proba})"
]
},
{
"cell_type": "markdown",
"metadata": {},
Expand Down Expand Up @@ -1844,7 +1702,7 @@
}
],
"source": [
"# Step 14: Generate requirements files\n",
"# Step 13: Generate requirements files\n",
"requirements_json = pzmm.JSONFiles.create_requirements_json(output_path)\n",
"\n",
"import json\n",
Expand All @@ -1865,7 +1723,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"Alright, we should have all our files now, so let's import our model and its files into SAS Model Manager! "
"Alright, we should have all our files now, so let's import our model and its files into SAS Model Manager! With the following function, we can automatically generate the score code and import the model into SAS Model Manager in one step."
]
},
{
Expand All @@ -1874,30 +1732,7 @@
"metadata": {},
"outputs": [],
"source": [
"# Step 15: Import model into SAS Model Manager \n",
"zipIOFile = pzmm.ZipModel.zip_files(model_files=output_path, model_prefix=model_prefix, is_viya4=True)\n",
"\n",
"mr.create_project(mm_project, 'Public')\n",
"\n",
"mr.import_model_from_zip(model_prefix, mm_project, zipIOFile)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"And now your model is available in SAS Model Manager where you can run the scoring test, update the model usage properties, add Key Performance Indicators (KPIs) thresholds, and run performance monitoring against the model. Managing the rest of the model lifecycle in SAS Model Manager will complete the rest of the model card.\n",
"\n",
"Instead of writing the score code manually and importing the model, you can run the following function to automatically generate the score code and import the model into SAS Model Manager in one step. Automatically generating the score code makes things easy, but manually writing allows greater customization and control. Only run the block below if you don't want to run steps 13 and 15. \n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Alternate to Step 13 and Step 15: Automatically generate score code and import model \n",
"# Step 14: Automatically generate score code and import model \n",
"pzmm.ImportModel.import_model(\n",
" model_files=output_path, # Where are the model files?\n",
" model_prefix=model_prefix, # What is the model name?\n",
Expand All @@ -1910,12 +1745,20 @@
" target_values=[\"0\", \"1\"], # What are the expected values of the target variable?\n",
" target_index=1, # What is the index of the target value in target_values?\n",
" model_file_name=model_prefix + \".pickle\", # How was the model file serialized?\n",
" missing_values=False # Does the data include missing values?\n",
" missing_values=False, # Does the data include missing values?\n",
" preprocess_function=preprocess_function # What do we want to do to the data before we score it?\n",
" )\n",
" # Reinitialize the score_code variable when writing more than one model's score code\n",
"pzmm.ScoreCode.score_code = \"\""
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"And now your model is available in SAS Model Manager where you can run the scoring test, update the model usage properties, add Key Performance Indicators (KPIs) thresholds, and run performance monitoring against the model. Managing the rest of the model lifecycle in SAS Model Manager will complete the rest of the model card."
]
},
{
"cell_type": "markdown",
"metadata": {},
Expand Down

0 comments on commit 2bd7d80

Please sign in to comment.