Predictions for fantasy baseball¶
Millions of people play fantasy baseball using leagues that are typically draft- or auction-based. Choosing a team based on your favorite players—or simply on last year's performance without any regard for regression to the mean—is likely to field a weaker team. Baseball is one of the most "documented" of all sports, statistics-wise. With the wealth of collected information, you can derive a better estimate of each player's true talent level and their likely performance in the coming year using machine learning. This allows for better drafting and also helps avoid overpaying for players coming off of "career" seasons.
When drafting players for fantasy baseball, you must make decisions based on the player's performance over their career to date, as well as effects like aging. Basing evaluation on personal interpretation of the player's performance is likely to cause you to overvalue a player's most recent performance. In other words, it's common to overvalue a player coming off a career year or undervalue a player coming off a bad year. The goal is to generate a better estimate of the player's value in the next year based on what he has done in prior years. If you build a machine learning model to predict a player's performance in the next year based on their previous performance, it will help you identify when these over- or under-performances are flukes versus when they are actual indicators of that player’s future performance.
To run this notebook on your own, use the download icon in the top right of this page.
Setup¶
This example assumes that the DataRobot Python client package has been installed and configured with the credentials of a DataRobot user with API access permissions. It also utilizes the pybaseball module, which scrapes data from Fangraphs.com, a popular baseball website with many statistics and a publicly available API. The following steps import the required libraries. In order to successfully use the pybaseball module, you must have Python 3.10+ installed in your environment.
Retrieve baseball data¶
This notebook uses the pybaseball module to get player seasons from 2012 through 2021. In this workflow, the machine learning algorithm learns patterns from pre-COVID era data as well as data from 2020 and 2021. This data should help show how well the top model is able to learn how to work around a shortened season.
Fangraphs provides more than 300 features about players each season, from the most superficial statistics like batting average (AVG) and home run counts (HR), to the most in-depth statistics like expected weighted on-base average (xWOBA) and barrel contact percentage (Barrel%). DataRobot sifts through many of these features to find the ones that signal future performance best.
For this notebook, the focus is on predicting a batting average in the following season. Batting average is a common statistic used in all fantasy baseball league types. It is a rate statistic, rather than a total like home runs, so you can leverage many rate statistic features of players and mostly avoid volume statistics outside of plate appearances (PA).
Import libraries¶
Use the pybaseball Github repo to install the package. You must have Python 3.10+ installed in your environment for this package to successfully install.
pip install git+https://github.com/jldbc/pybaseball
import datetime
from datetime import datetime
import datarobot as dr
import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
import numpy as np
import pandas as pd
Import player batting data¶
Starting in 2007, Pitch Info Solutions started tracking some advanced pitch-by-pitch metrics for batters and pitchers, making it a good year to start the model training.
Download the data in two chunks:
batters_training_2007_2013 = pd.read_csv(
"https://docs.datarobot.com/en/docs/api/guide/common-case/batters_training_2007_2013.csv"
)
batters_training_2014_2021 = pd.read_csv(
"https://docs.datarobot.com/en/docs/api/guide/common-case/batters_training_2014_2021.csv"
)
batters_training = pd.concat([batters_training_2007_2013, batters_training_2014_2021], axis=0)
batters_training.shape
(14118, 319)
Data preparation¶
The snippets below create a "primary" target dataset, as well as a secondary dataset from which you derive features on each player's past performance across many statistics. Note that you may need to run the cell below multiple times.
batters_training["Date"] = [str(year) + "-03-01" for year in batters_training["Season"]]
batters_training["COVID"] = [1 if year == 2020 else 0 for year in batters_training["Season"]]
batters_training = batters_training[
[ # Basic info
"IDfg",
"Season",
"Name",
"Date",
"COVID",
"Age",
"PA",
# Superficial rates
"BB%",
"K%",
"AVG",
# Expected rates based on batted ball data
"xBA",
"Barrel%",
"maxEV",
"xSLG",
"Oppo%",
"HR/FB",
"IFFB%",
# Plate discipline
"CStr%",
"SwStr%",
"O-Swing% (pi)",
"Z-Swing% (pi)",
"O-Contact% (pi)",
"Z-Contact% (pi)",
"Zone% (pi)",
# Speed + position adjustment
"SB",
"CS",
"BsR",
"Pos",
]
]
batters_training["SB%"] = batters_training["SB"] / (batters_training["SB"] + batters_training["CS"])
batters_training["SB/PA"] = batters_training["SB"] / (batters_training["PA"])
batters_training["BsR/PA"] = batters_training["BsR"] / (batters_training["PA"])
# Scale plate appearances (PA) to the season's highest total, to better adjust for COVID-shortened season
max_pa = pd.DataFrame(batters_training.groupby("Season")["PA"].max())
max_pa.rename(columns={"PA": "PA_max"}, inplace=True)
batters_training = pd.merge(batters_training, max_pa, how="left", on="Season")
batters_training["PA_scaled"] = batters_training["PA"] / batters_training["PA_max"]
# Use a weight so that the model can learn more from players with greater sample size
batters_training["weight"] = np.sqrt(batters_training["PA"])
# Create primary and secondary datasets for Feature Discovery (automated feature engineering)
batters_primary = batters_training[batters_training["Season"] > 2007][
["IDfg", "Date", "COVID", "Age", "weight", "AVG"]
]
batters_secondary = batters_training[
[
"IDfg",
"Date",
"COVID",
"PA_scaled",
"BB%",
"K%",
"AVG",
"xBA",
"Barrel%",
"maxEV",
"xSLG",
"Oppo%",
"HR/FB",
"IFFB%",
"CStr%",
"SwStr%",
"O-Swing% (pi)",
"Z-Swing% (pi)",
"O-Contact% (pi)",
"Z-Contact% (pi)",
"Zone% (pi)",
"SB%",
"SB/PA",
"BsR/PA",
"Pos",
]
]
Connect to DataRobot¶
Read more about different options for connecting to DataRobot from the Python client.
# If the config file is not in the default location described in the API Quickstart guide, '~/.config/datarobot/drconfig.yaml', then you will need to call
# dr.Client(config_path='path-to-drconfig.yaml')
Modeling with Feature Discovery¶
Create a DataRobot project to train many models against the assembled dataset. Then you will leverage an automated feature engineering tool in DataRobot called Feature Discovery, which will use the secondary dataset created above to derive rolling, time-aware features about players' recent performance history. In this example, DataRobot looks back up to three seasons to derive a player's recent performance, though it's worth testing longer and shorter feature derivation windows.
Define projects settings¶
# Register secondary dataset in AI Catalog
secondary_dataset = dr.Dataset.create_from_in_memory_data(batters_secondary)
# Define relationships between datasets and lookback "feature derivation" windows
dataset_definitions = [
{
"identifier": "BA_secondary_3years",
"catalogVersionId": secondary_dataset.version_id,
"catalogId": secondary_dataset.id,
# "feature_list_id": secondary_dataset_featurelist.id, # default is Informative features
"primaryTemporalKey": "Date",
"snapshotPolicy": "latest",
},
{
"identifier": "BA_secondary_2",
"catalogVersionId": secondary_dataset.version_id,
"catalogId": secondary_dataset.id,
# "feature_list_id": secondary_dataset_featurelist.id, # default is Informative features
"primaryTemporalKey": "Date",
"snapshotPolicy": "latest",
},
{
"identifier": "BA_secondary_3",
"catalogVersionId": secondary_dataset.version_id,
"catalogId": secondary_dataset.id,
# "feature_list_id": secondary_dataset_featurelist.id, # default is Informative features
"primaryTemporalKey": "Date",
"snapshotPolicy": "latest",
},
]
relationships = [
{
"dataset2Identifier": "BA_secondary_3years",
"dataset1Keys": ["IDfg"],
"dataset2Keys": ["IDfg"],
"featureDerivationWindowStart": -3,
"featureDerivationWindowEnd": 0,
"featureDerivationWindowTimeUnit": "YEAR",
"predictionPointRounding": 1,
"predictionPointRoundingTimeUnit": "SECOND",
},
{
"dataset2Identifier": "BA_secondary_2",
"dataset1Keys": ["IDfg"],
"dataset2Keys": ["IDfg"],
"featureDerivationWindowStart": -2,
"featureDerivationWindowEnd": -1,
"featureDerivationWindowTimeUnit": "YEAR",
"predictionPointRounding": 1,
"predictionPointRoundingTimeUnit": "SECOND",
},
{
"dataset2Identifier": "BA_secondary_3",
"dataset1Keys": ["IDfg"],
"dataset2Keys": ["IDfg"],
"featureDerivationWindowStart": -3,
"featureDerivationWindowEnd": -2,
"featureDerivationWindowTimeUnit": "YEAR",
"predictionPointRounding": 1,
"predictionPointRoundingTimeUnit": "SECOND",
},
]
relationship_config = dr.RelationshipsConfiguration.create(
dataset_definitions=dataset_definitions, relationships=relationships
)
# Set up specific backtest segments
# Set holdout partition full season 2021
# The 2020 season has been set with the date 2020-03-01, so start this partition on 2020-03-02
# set gap of 1 year to avoid validating on 2020 (though 2020 will still be used in final model training)
partitioning_spec = dr.DatetimePartitioningSpecification(
datetime_partition_column="Date",
gap_duration="P1Y",
holdout_start_date=datetime(2020, 3, 2),
holdout_duration="P1Y",
number_of_backtests=2,
use_time_series=False,
)
# Set the backtest partition--used to hyperparameter tune the models--to the non-COVID 2019 season
partitioning_spec.backtests = [
dr.BacktestSpecification(
0, # backtest 1 index
gap_duration="P0Y",
validation_start_date=datetime(2018, 3, 2),
validation_duration="P1Y",
),
dr.BacktestSpecification(
1, # backtest 2 index
gap_duration="P0Y",
validation_start_date=datetime(2017, 3, 2),
validation_duration="P1Y",
),
# Input other backtest periods below if desired
# dr.BacktestSpecification(
# 1, # backtest 2 index
# gap_duration = 'P0Y',
# validation_start_date=datetime(2018,3,2),
# validation_duration = 'P1Y')
]
# Create advanced options object with weights
advanced_options = dr.AdvancedOptions(weights="weight", consider_blenders_in_recommendation=True)
Use the snippet below to get a full explanation of the advanced options configured for the notebook.
?dr.AdvancedOptions
Create project¶
EXISTING_PROJECT_ID = None # If you've already created a project, replace None with the ID here
if EXISTING_PROJECT_ID is None:
# Create project and pass in data
project = dr.Project.create(sourcedata = batters_primary,
project_name = 'Batting Average Predictions for Fantasy Baseball: 2007 - 2021'
dataset_filename="batters_primary")
# Set the project target to the appropriate feature. Use the LogLoss metric to measure performance
project.set_target(target='broken',
mode=dr.AUTOPILOT_MODE.QUICK,
worker_count='-1')
else:
# Fetch the existing project
project = dr.Project.get(EXISTING_PROJECT_ID)
project.wait_for_autopilot(check_interval=30)
# The featurelist from the primary dataset effectively includes "known in advance" features
# All features from the secondary dataset will be lagged
primary_featurelist = project.create_featurelist(
"Primary model features",
['COVID','Age'])
Initiate Autopilot¶
# Start project
project.analyze_and_model(
target="AVG",
worker_count=-1,
advanced_options=advanced_options,
partitioning_method=partitioning_spec,
featurelist_id=primary_featurelist.id,
relationships_configuration_id=relationship_config.id,
max_wait=3600, # allow up to an hour for pre-processing before timeout
)
View project in UI¶
If you want to view the project in the DataRobot UI, use the following snippet to retrieve the project's URL and use it to navigate to the application.
# Get project URL
project_url = project.get_uri()
print(project_url)
https://app.datarobot.com/projects/62ec4628387ba15033da9153/models
project.wait_for_autopilot(check_interval=60)
In progress: 0, queued: 0 (waited: 0s)
Evaluate model performance¶
In order to measure model performance, first select the top-performing model based on a specific performance metric (i.e., LogLoss) and then evaluate several different types of charts, such as Lift Chart, ROC Curve, and Feature Importance. There are two helper functions that you need to build in order to simplify producing these model insights.
Because this project utilized time-based partitioning, there are not out-of-sample validation metrics for models fit to 100% of the data (including the recommended model). Investigate the parent of the recommended model, fit to the training data, using the following code. You can reference more information about model evaluation tools here.
def sorted_by_metric(models, test_set, metric):
models_with_score = [model for model in models if model.metrics[metric][test_set] is not None]
return sorted(models_with_score, key=lambda model: model.metrics[metric][test_set])
# Unlock holdout and get list of models trained
project.unlock_holdout()
models = project.get_models()
# Uncomment if this is not set when creating the project
metric = project.metric
# Get the top-performing model
model_top = project.get_top_model(metric=metric)
print(
"""The top performing model is {model} using metric, {metric}""".format(
model=str(model_top), metric=metric
)
)
The top performing model is Model('AVG Blender') using metric, Weighted RMSE
# Set styling
dr_dark_blue = "#08233F"
dr_blue = "#1F77B4"
dr_orange = "#FF7F0E"
dr_red = "#BE3C28"
dr_light_blue = "#3CA3E8"
# Histogram function
def rebin_df(raw_df, number_of_bins):
cols = ["bin", "actual_mean", "predicted_mean", "bin_weight"]
new_df = pd.DataFrame(columns=cols)
current_prediction_total = 0
current_actual_total = 0
current_row_total = 0
x_index = 1
bin_size = 60 / number_of_bins
for rowId, data in raw_df.iterrows():
current_prediction_total += data["predicted"] * data["bin_weight"]
current_actual_total += data["actual"] * data["bin_weight"]
current_row_total += data["bin_weight"]
if (rowId + 1) % bin_size == 0:
x_index += 1
bin_properties = {
"bin": ((round(rowId + 1) / 60) * number_of_bins),
"actual_mean": current_actual_total / current_row_total,
"predicted_mean": current_prediction_total / current_row_total,
"bin_weight": current_row_total,
}
new_df = new_df.append(bin_properties, ignore_index=True)
current_prediction_total = 0
current_actual_total = 0
current_row_total = 0
return new_df
def matpltlib_residual(residual):
if residual:
validation = residuals[2]
actual, predicted, _, _ = zip(*validation.data)
data = {"actual": actual, "predicted": predicted}
data_frame = pd.DataFrame(data)
plot = data_frame.plot.scatter(
x="actual",
y="predicted",
legend=False,
color=dr_light_blue,
)
plot.set_facecolor(dr_dark_blue)
# define our axes with a minuscule bit of padding
min_x = min(data["actual"]) - 5
max_x = max(data["actual"]) + 5
min_y = min(data["predicted"]) - 5
max_y = max(data["predicted"]) + 5
biggest_value = max(abs(i) for i in [min_x, max_x, min_y, max_y])
# plot a diagonal 1:1 line to show the "perfect fit" case
diagonal = np.linspace(-biggest_value, biggest_value, 100)
plt.plot(diagonal, diagonal, color="gray")
plt.xlabel("Actual Value")
plt.ylabel("Predicted Value")
plt.axis("equal")
plt.xlim(min_x, max_x)
plt.ylim(min_y, max_y)
plt.title("Predicted Values vs. Actual Values", y=1.04)
Lift Chart¶
The Lift Chart shows you how close model predictions are to the actual values of the target in the training data. Lift Chart data includes the average predicted value and the average actual values of the target, sorted by the prediction values in ascending order and split into up to bins. You can select 5, 10, 15, 20, 30, or 60 bins.
The predictions shown below were generated on 2021 data (the holdout data), using a model that learned patterns from the 2012 - 2018 seasons about how a player's recent performance correlates to their next-season performance. 2021 is a tough year to predict because the season immediately prior was a COVID season. Still, the model did admirably well.
# Lift chart function
def matplotlib_lift(bins_df, bin_count, ax):
grouped = rebin_df(bins_df, bin_count)
ax.plot(
range(1, len(grouped) + 1),
grouped["predicted_mean"],
marker="+",
lw=1,
color=dr_blue,
label="predicted",
)
ax.plot(
range(1, len(grouped) + 1),
grouped["actual_mean"],
marker="*",
lw=1,
color=dr_orange,
label="actual",
)
ax.set_xlim([0, len(grouped) + 1])
ax.set_facecolor(dr_dark_blue)
ax.legend(loc="best")
ax.set_title("Lift chart {} bins".format(bin_count))
ax.set_xlabel("Sorted Prediction")
ax.set_ylabel("Value")
return grouped
lift_chart = model_top.get_lift_chart("holdout")
# Save the result into a pandas dataframe
lift_df = pd.DataFrame(lift_chart.bins)
bin_counts = [10, 30]
f, axarr = plt.subplots(len(bin_counts))
f.set_size_inches((8, 4 * len(bin_counts)))
rebinned_dfs = []
for i in range(len(bin_counts)):
rebinned_dfs.append(matplotlib_lift(lift_df, bin_counts[i], axarr[i]))
plt.tight_layout()
# plt.show()
Feature Impact¶
Feature Impact measures how important a feature is in the context of a model. It measures how much the accuracy of a model would decrease if that feature was removed.
Feature Impact is available for all model types and works by altering input data and observing the effect on a model’s score. It is an on-demand feature, meaning that you must initiate a calculation to see the results. Once DataRobot computes the feature impact for a model, that information is saved with the project.
feature_impacts = model_top.get_or_request_feature_impact()
# Limit size to make the chart display better
# Displays the top 25 values
if len(feature_impacts) > 25:
feature_impacts = feature_impacts[0:24]
# Formats the ticks from a float into a percent
percent_tick_fmt = mtick.PercentFormatter(xmax=1.0)
impact_df = pd.DataFrame(feature_impacts)
impact_df.sort_values(by="impactNormalized", ascending=True, inplace=True)
# Positive values are blue, negative are red
bar_colors = impact_df.impactNormalized.apply(lambda x: dr_red if x < 0 else dr_blue)
ax = impact_df.plot.barh(
x="featureName", y="impactNormalized", legend=False, color=bar_colors, figsize=(10, 8)
)
ax.xaxis.set_major_formatter(percent_tick_fmt)
ax.xaxis.set_tick_params(labeltop=True)
ax.xaxis.grid(True, alpha=0.2)
ax.set_facecolor(dr_dark_blue)
plt.ylabel("")
plt.xlabel("Effect")
plt.xlim((None, 1)) # Allow for negative impact
plt.title("Feature Impact", y=1.04);
Make predictions¶
Test predictions¶
After determining the top-performing model from the Leaderboard, upload a prediction test dataset to verify that the model generates predictions successfully before deploying the model to a production environment. The predictions are returned as a Pandas dataframe. Note that this example uses two methods for creating the scoring dataset. In most cases, you'd be running this in the offseason, and you wouldn't have a definitive list of players for the following season. In that case, the second option below makes sense—taking the players from the prior season (2021) and creating a scoring dataset for 2022.
# Get recommended model
rec_model = project.recommended_model()
# Option 1: Scoring data using the current (2022) season of players (if the season has commenced)
scoring_data_2022 = batting_stats(2022, 2022, qual=1, split_seasons=True)
scoring_data_2022["Date"] = [str(year) + "-03-01" for year in scoring_data_2022["Season"]]
scoring_data_2022["COVID"] = [1 if year == 2020 else 0 for year in scoring_data_2022["Season"]]
scoring_data_2022 = scoring_data_2022[["IDfg", "Name", "Date", "COVID", "Age", "PA", "AVG"]]
scoring_data_2022["weight"] = np.sqrt(scoring_data_2022["PA"])
# Option 2: Scoring data created from the prior (2021) season of players using training data
scoring_data_2022_from_training = batters_training[batters_training["Date"] == "2021-03-01"]
scoring_data_2022_from_training = scoring_data_2022_from_training[
["IDfg", "Date", "COVID", "Age", "PA", "weight", "AVG"]
]
scoring_data_2022_from_training["Age"] = scoring_data_2022_from_training["Age"] + 1
scoring_data_2022_from_training["Date"] = [
str(int(date[0:4]) + 1) + "-03-01" for date in scoring_data_2022_from_training["Date"]
]
scoring_data_2022.to_csv("2022_AVG_scoring_data.csv")
scoring_data_2022
prediction_dataset = project.upload_dataset(scoring_data_2022)
predict_job = rec_model.request_predictions(prediction_dataset.id)
predictions = predict_job.get_result_when_complete()
pd.concat([scoring_data_2022, predictions], axis=1)
predictions.positive_probability.plot(kind="hist", title="Predicted Probabilities")
| IDfg | Name | Date | COVID | Age | PA | AVG | weight | row_id | prediction | |
|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 19952 | Khalil Lee | 2022-03-01 | 0 | 24 | 2 | 0.5 | 1.414214 | 0 | 0.254665 |
| 1 | 21635 | Nick Maton | 2022-03-01 | 0 | 25 | 7 | 0.5 | 2.645751 | 1 | 0.258107 |
| 2 | 24770 | James Outman | 2022-03-01 | 0 | 25 | 12 | 0.6 | 3.464102 | 2 | 0.275660 |
| 3 | 17734 | Evan Phillips | 2022-03-01 | 0 | 27 | 1 | 0.0 | 1.000000 | 3 | 0.295481 |
| 4 | 20178 | Miguel Vargas | 2022-03-01 | 0 | 22 | 4 | 0.5 | 2.000000 | 4 | 0.245772 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 620 | 20510 | Jack Larsen | 2022-03-01 | 0 | 27 | 1 | 0.0 | 1.000000 | 620 | 0.274557 |
| 621 | 21840 | Freddy Fermin | 2022-03-01 | 0 | 27 | 7 | 0.0 | 2.645751 | 621 | 0.210788 |
| 622 | 21853 | Will Benson | 2022-03-01 | 0 | 24 | 2 | 0.0 | 1.414214 | 622 | 0.241940 |
| 623 | 22254 | Joe Perez | 2022-03-01 | 0 | 22 | 1 | 0.0 | 1.000000 | 623 | 0.225691 |
| 624 | 23378 | Jose Barrero | 2022-03-01 | 0 | 24 | 3 | 0.0 | 1.732051 | 624 | 0.262743 |
625 rows × 10 columns
Deploy a model to production¶
If you are happy with the model's performance, you can deploy it to a production environment with MLOps. Deploying the model will free up workers, as data scored through the deployment doesn't use any modeling workers. Furthermore, you are no longer restricted on the amount of data to score; score over 100GB with the deployment. Deployments also offer many model management benefits: monitoring service, data drift, model comparison, retraining, and more.
# Create a prediction server
prediction_server = dr.PredictionServer.list()[0]
# Get the top-performing model. Uncomment if this did not execute in the previous cell
# rec_model = project.recommended_model()
deployment = dr.Deployment.create_from_learning_model(
rec_model.id,
label="Predictions for Fantasy Baseball",
description="Predictions for Fantasy Baseball",
default_prediction_server_id=prediction_server.id,
)
deployment.id
Deployment(Predictions for Fantasy Baseball)
Configure batch predictions¶
After the model has been deployed, DataRobot creates an endpoint for real-time scoring. The deployment allows you to use DataRobot's batch prediction API to score large datasets with a deployed DataRobot model.
The batch prediction API provides flexible intake and output options when scoring large datasets using prediction servers. The API is exposed through the DataRobot Public API and can be consumed using a REST-enabled client or Public API bindings for DataRobot's Python client.
Set the deployment ID¶
Before proceeding, provide the deployed model's deployment ID (which can be retrieved from the deployment's Overview tab), commented out below. If you created the deployment through the API, you can get the ID from that deployment object.
deployment_id = deployment.id
# deployment_id = "YOUR_DEPLOYMENT_ID"
Determine input and output options¶
DataRobot's batch prediction API allows you to score data from and to multiple sources. You can take advantage of the credentials and data sources you have already established previously through the UI for easy scoring. Credentials are usernames and passwords, while data sources are any databases with which you have previously established a connection (e.g., Snowflake). View the example code below outlining how to query credentials and data sources.
You can reference the full list of DataRobot's supported input and output options.
The snippet below shows how you can query all credentials tied to a DataRobot account.
dr.Credential.list()
[Credential('60bfeedd23791e786c9e0eab', 'github-application-oauth', 'oauth'),
Credential('6169fe16789e06406eec9716', 'Snowflake CFDS', 'basic'),
Credential('62cf3cbb3dbe207eb1f0b0a3', 'Snowflake SANDBOX', 'basic')]
The output above returns multiple sets of credentials. The alphanumeric string included in each item of the list is the credentials ID. You can use that ID to access credentials through the API.
The snippet below shows how you can query all data sources tied to a DataRobot account. The second line lists each datastore with an alphanumeric string; that is the datastore ID.
dr.DataStore.list()
print(dr.DataStore.list()[0].id)
6169fdff337affb9a4d22c98
Scoring examples¶
The snippets below demonstrate how to score data with the Batch Prediction API. Edit the intake_settings and output_settings to suit your needs. You can mix and match until you get the outcome you prefer.
Score from CSV or Pandas dataframe to CSV¶
This notebook uses this approach, but see other examples below depending on your setup.
# Scoring without Prediction Explanations
results_df = deployment.predict_batch(source=Path("scoring_data_2022"))
result_df
# Scoring with Prediction Explanations
# # Note: this will not work on a model with OTV partitioning trained to 100% of the training data
# job = dr.BatchPredictionJob.score(
# deployment=deployment.id,
# intake_settings={
# 'type': 'localFile',
# 'file': scoring_data_2022,
# },
# output_settings={
# 'type': 'localFile',
# 'path': '2022_fantasy_predictions.csv',
# },
# max_explanations=3 #Compute Prediction Explanations for the amount of features indicated here
# )
Score from S3 to S3¶
# dr.BatchPredictionJob.score(
# deployment_id,
# intake_settings={
# 'type': 's3',
# 'url': 's3://', #Provide the URL of your datastore here
# 'credential_id': 'YOUR_CREDENTIAL_ID_FROM_ABOVE', # Provide your credentials here
# },
# output_settings={
# 'type': 's3',
# 'url': 's3://theos-test-bucket/lending_club_scored2.csv',
# 'credential_id': 'YOUR_CREDENTIAL_ID_FROM_ABOVE'
# }
# )
Score from JDBC to JDBC¶
# dr.BatchPredictionJob.score(
# deployment_id,
# intake_settings = {
# 'type': 'jdbc',
# 'table': 'table_name',
# 'schema': 'public',
# 'dataStoreId': data_store.id, #Provide the ID of your datastore here
# 'credentialId': cred.credential_id # Provide your credentials here
# },
# output_settings = {
# 'type': 'jdbc',
# 'table': 'table_name',
# 'schema': 'public',
# 'statementType': 'insert',
# 'dataStoreId': data_store.id,
# 'credentialId': cred.credential_id
# }
# )
Evaluate predictions¶
Take a look at some of the players with the best predictions for 2022 (AVG_PREDICTION), and how they are actually doing so far (AVG). Next, create the underlying table for a lift chart on players with at least 250 plate appearances (PA) to evaluate the model. Unsuprisingly, the model's predictions in extreme buckets—especially on the low end—are a little closer to the overall mean than the actual batting averages. This is likely because this season is only two-thirds complete, allowing more variability in the actual outcomes than the model was trained on.
# View top players by batting average prediction vs. actual
predictions = pd.read_csv("2022_fantasy_predictions.csv")
predictions = pd.concat([scoring_data_2022, predictions], axis=1)
predictions[predictions["PA"] > 250].sort_values("AVG_PREDICTION", ascending=False).head(20)
| IDfg | Name | Date | COVID | Age | PA | AVG | weight | AVG_PREDICTION | DEPLOYMENT_APPROVAL_STATUS | |
|---|---|---|---|---|---|---|---|---|---|---|
| 37 | 17982 | Ty France | 2022-03-01 | 0 | 27 | 388 | 0.308 | 19.697716 | 0.290818 | APPROVED |
| 10 | 15640 | Aaron Judge | 2022-03-01 | 0 | 30 | 450 | 0.298 | 21.213203 | 0.285348 | APPROVED |
| 15 | 11579 | Bryce Harper | 2022-03-01 | 0 | 29 | 275 | 0.318 | 16.583124 | 0.283950 | APPROVED |
| 50 | 3473 | Anthony Rizzo | 2022-03-01 | 0 | 32 | 415 | 0.227 | 20.371549 | 0.283094 | APPROVED |
| 103 | 19326 | Bryan Reynolds | 2022-03-01 | 0 | 27 | 365 | 0.255 | 19.104973 | 0.282391 | APPROVED |
| 98 | 12916 | Francisco Lindor | 2022-03-01 | 0 | 28 | 455 | 0.262 | 21.330729 | 0.281507 | APPROVED |
| 212 | 19290 | Randy Arozarena | 2022-03-01 | 0 | 27 | 417 | 0.251 | 20.420578 | 0.279437 | APPROVED |
| 200 | 23792 | Alek Thomas | 2022-03-01 | 0 | 22 | 271 | 0.257 | 16.462078 | 0.276552 | APPROVED |
| 144 | 7859 | Charlie Blackmon | 2022-03-01 | 0 | 35 | 417 | 0.266 | 20.420578 | 0.276427 | APPROVED |
| 12 | 9218 | Paul Goldschmidt | 2022-03-01 | 0 | 34 | 422 | 0.332 | 20.542639 | 0.275934 | APPROVED |
| 14 | 17350 | Rafael Devers | 2022-03-01 | 0 | 25 | 388 | 0.322 | 19.697716 | 0.274905 | APPROVED |
| 359 | 12979 | Javier Baez | 2022-03-01 | 0 | 29 | 373 | 0.221 | 19.313208 | 0.274410 | APPROVED |
| 36 | 22581 | Alejandro Kirk | 2022-03-01 | 0 | 23 | 346 | 0.305 | 18.601075 | 0.274066 | APPROVED |
| 46 | 11615 | Brandon Drury | 2022-03-01 | 0 | 29 | 389 | 0.275 | 19.723083 | 0.272839 | APPROVED |
| 62 | 14161 | Byron Buxton | 2022-03-01 | 0 | 28 | 328 | 0.221 | 18.110770 | 0.271382 | APPROVED |
| 16 | 18360 | Austin Riley | 2022-03-01 | 0 | 25 | 445 | 0.300 | 21.095023 | 0.271287 | APPROVED |
| 140 | 12552 | Eugenio Suarez | 2022-03-01 | 0 | 30 | 445 | 0.234 | 21.095023 | 0.270989 | APPROVED |
| 155 | 18401 | Ronald Acuna Jr. | 2022-03-01 | 0 | 24 | 315 | 0.257 | 17.748239 | 0.270231 | APPROVED |
| 82 | 15362 | Jeff McNeil | 2022-03-01 | 0 | 30 | 352 | 0.302 | 18.761663 | 0.269978 | APPROVED |
| 109 | 12927 | Brandon Nimmo | 2022-03-01 | 0 | 29 | 428 | 0.272 | 20.688161 | 0.269945 | APPROVED |
Compare predictions to actual values¶
predictions["predicted_bucket"] = pd.qcut(predictions["AVG_PREDICTION"], 10)
predictions.sort_values("AVG_PREDICTION", ascending=False).groupby("predicted_bucket").agg(
{"AVG": ["mean", "count"], "PA": "mean"}
)
| AVG | PA | ||
|---|---|---|---|
| mean | count | mean | |
| predicted_bucket | |||
| (0.10099999999999999, 0.2] | 0.198841 | 63 | 165.174603 |
| (0.2, 0.214] | 0.200563 | 80 | 171.062500 |
| (0.214, 0.218] | 0.200667 | 45 | 169.155556 |
| (0.218, 0.223] | 0.206387 | 62 | 175.758065 |
| (0.223, 0.231] | 0.209349 | 63 | 197.920635 |
| (0.231, 0.237] | 0.221500 | 62 | 205.548387 |
| (0.237, 0.243] | 0.197597 | 62 | 176.564516 |
| (0.243, 0.251] | 0.224778 | 63 | 205.746032 |
| (0.251, 0.263] | 0.240548 | 62 | 204.241935 |
| (0.263, 0.295] | 0.262714 | 63 | 212.857143 |