ai_cdss.DataProcessor

class ai_cdss.DataProcessor(weights=[1, 1, 1], alpha=0.5, context=None)

A class for processing data and computing a final weighted score.

The final score is computed as:

\[S = \alpha \cdot A + \beta \cdot DM + \gamma \cdot PPF\]

where:

• \(A\) is Adherence

• \(DM\) is the Difficulty Modulator

• \(PPF\) is the Patient Prescription Factor

Parameters:

weightsList[float]

List of weights \(\alpha\), \(\beta\), \(\gamma\) for computing the final score.

alphafloat

The smoothing factor for EWMA, controlling how much past values influence the trend.

Methods

build_prescription_days(session_df[, ...])	This feature builder must return active prescriptions signaled as: PRESCRIPTION_ENDING_DATE == 2100-01-01 00:00:00
build_recent_adherence(session_df)	This feature builder must return adherence for patient protocols, with the following considerations: - Adherence computed as SESSION_TIME / PRESCRIBED_SESSION_TIME - Adherence is considered 0 if presribed session is not performed - BUT is not considered if no session was performed that day.
build_usage(session_df)	This feature builder must return how many times protocols are used in this format: PATIENT_ID PROTOCOL_ID USAGE 12 220 2 12 231 1 12 233 0
build_week_usage(session_df[, scoring_date])	This feature builder must return how many times protocols are used in this week in this format: PATIENT_ID PROTOCOL_ID USAGE_WEEK 12 220 2 12 231 1 12 233 0
compute_score(data, protocol_metrics)	Initializes metrics based on legacy data and computes score
process_data(session_data, timeseries_data, ...)	Process and score patient-protocol combinations using session, timeseries, and PPF data.

build_delta_dm

process_data(session_data, timeseries_data, ppf_data, init_data)

Process and score patient-protocol combinations using session, timeseries, and PPF data. When bootstrapping patient, score is based on PPF only. Otherwise a weighted sum of DELTA_DM, ADHERENCE_RECENT and PPF determines the score.

Return type:

DataFrame[ScoringSchema]

Parameters:

session_dataDataFrame[SessionSchema]

Session-level data including adherence and scheduling information.

timeseries_dataDataFrame[TimeseriesSchema]

Timepoint-level data including DMs and performance metrics.

ppf_dataDataFrame[PPFSchema]

Patient-protocol fitness values and contributions.

Returns:

DataFrame[ScoringSchema]

Final scored dataframe with protocol recommendations.

compute_score(data, protocol_metrics)

Initializes metrics based on legacy data and computes score

Parameters:

datapd.DataFrame

Aggregated session and timeseries data per patient-protocol.

Returns:

pd.DataFrame

Scored DataFrame sorted by patient and protocol.

build_recent_adherence(session_df)

This feature builder must return adherence for patient protocols, with the following considerations: - Adherence computed as SESSION_TIME / PRESCRIBED_SESSION_TIME - Adherence is considered 0 if presribed session is not performed

• BUT is not considered if no session was performed that day.

Additionally a recency bias is applied.

Return type:

DataFrame

build_usage(session_df)

This feature builder must return how many times protocols are used in this format: PATIENT_ID PROTOCOL_ID USAGE 12 220 2 12 231 1 12 233 0

Protocols that have not been yet prescribed for a patient are not returned.

Return type:

DataFrame

build_week_usage(session_df, scoring_date=None)

This feature builder must return how many times protocols are used in this week in this format: PATIENT_ID PROTOCOL_ID USAGE_WEEK 12 220 2 12 231 1 12 233 0

Protocols that have not been yet prescribed for a patient are not returned.

Return type:

DataFrame

build_prescription_days(session_df, scoring_date=None)

This feature builder must return active prescriptions signaled as: PRESCRIPTION_ENDING_DATE == 2100-01-01 00:00:00

In the following format, (encoding weekdays from 0-6): PATIENT_ID PROTOCOL_ID DAYS 12 220 [2] 12 233 [0]

Return type:

DataFrame

Examples using ai_cdss.DataProcessor

Basic usage of CDSS to generate a protocol recommendation.

Basic usage of CDSS to generate a protocol recommendation.