Skip to content

Commit

Permalink
Merge branch 'main' into 22631-fix-malformed-training-indices
Browse files Browse the repository at this point in the history
  • Loading branch information
@fulpm
fulpm committed Jan 29, 2025
2 parents e22b1fb + a24cd0a commit c967371
Show file tree
Hide file tree
Showing 30 changed files with 1,369 additions and 691 deletions.
1 change: 1 addition & 0 deletions howso.amlg
View file
Original file line number Diff line number Diff line change
Expand Up @@ -235,6 +235,7 @@
"react_group"
"react_aggregate"
"react_series"
"react_series_stationary"
"react_series_utilities"
"react_utilities"
"remove_cases"
Expand Down
206 changes: 102 additions & 104 deletions howso/derive_utilities.amlg
View file
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@
rand_seed rand_seed

desired_conviction desired_conviction
use_regional_model_residuals use_regional_model_residuals
use_regional_residuals use_regional_residuals
feature_bounds_map
(if modified_feature_bounds_map
modified_feature_bounds_map
Expand All @@ -115,6 +115,7 @@
ordered_by_specified_features ordered_by_specified_features
generate_new_cases "no"
preserve_feature_values preserve_feature_values
holdout_queries holdout_queries
))
))

Expand Down Expand Up @@ -331,16 +332,6 @@
)

;automatically prepare inputs for BatchReactSeries for a time series model and outputs an updated assoc of react parameters.
;Ouputs null if no prep is needed and a string message if there was a validation error. On success, outputs:
; (assoc
; "action_features" action_features - list of features
; "initial_features" initial_features - list of features
; "initial_values" initial_values - list of list of values
; "derived_action_features" derived_action_features - list of features
; "derived_context_features" derived_context_features - list of features
; "series_stop_maps" series_stop_maps - list of assocs
; )
;
;parameters:
; init_time_steps: optional, time step values at which to begin synthesis
; final_time_steps: optional, time step values at which to end synthesis
Expand All @@ -357,10 +348,6 @@
; If "fixed", tracks the particular relevant series ID.
; If "dynamic", tracks the particular series ID, but is allowed to change the series ID that it tracks based on its current context.
; If "no", does not track any particular series ID.
; continue_series_values: optional, list of lists of values, when specified will continue this specific untrained series as defined by these values.
; continue_series flag will be ignored and treated as true if this value is specified.
; continue_series_features: optional, list of features corresponding to the values in each row of continue_series_values.
; This value is ignored if continue_series_values is not specified.
#!PrepTimeSeriesFeatures
(declare
(assoc single_series (false) )
Expand All @@ -381,7 +368,7 @@
)

(declare (assoc
all_contexts (append context_features series_context_features)
all_contexts series_context_features
n_samples
(if (size series_context_values)
(size series_context_values)
Expand All @@ -393,114 +380,124 @@
)
))

;The default initial features/values set the `start` feature to true.
(if (= 0 (size initial_features) (size initial_values))
;if all id features should be new, we do not need to retain IDs
(if output_new_series_ids
;if generating new series ids or series ids are given, only initial series progress of zero is needed
(if (or output_new_series_ids (size series_id_values))
(assign (assoc
initial_features (list ".series_progress")
initial_values (list (list 0.0))
))

;else if we want to retain the existing IDs, we must specify them as
;initial features (and matching initial values).
(let
(assoc
all_contexts_set (zip all_contexts)
series_id_features (list)
)

(assign (assoc
initial_features (list ".series_progress")
initial_values (list (list 0.0))
;Only ID features which are not already specified as contexts should be in initial values
series_id_features
(filter
(lambda (not (contains_index all_contexts_set (current_value)) ) )
(get !tsModelFeaturesMap "series_id_features")
)
))

;else if we want to retain the existing IDs, we must specify them as
;initial features (and matching initial values).
(let
(assoc
all_contexts_set (zip all_contexts)
series_id_features (list)
)

(assign (assoc
;Only ID features which are not already specified as contexts should be in initial values
series_id_features
(filter
(lambda (not (contains_index all_contexts_set (current_value)) ) )
(get !tsModelFeaturesMap "series_id_features")
)
))
(declare (assoc
all_id_values_defined
;true if all series_id_features are in series_context_features
(=
(size series_id_features)
(size (keep (zip series_context_features) series_id_features))
)
))

;ID features may be in one of the context lists
;(series_context or context features). It's possible that no IDs actually need to be set as initial.
(if (size series_id_features)
(let
(assoc
initial_values_list
;if continuing an untrained series, use the id values from the specified data
(if (size continue_series_values)
(map
(lambda
;only keep the series id values
(unzip
;create an assoc of features -> values from first row of series
(zip continue_series_features (first (current_value)))
series_id_features
)
;ID features may be in one of the context lists
;(series_context or context features). It's possible that no IDs actually need to be set as initial.
(if (size series_id_features)
(let
(assoc
initial_values_list
;if continuing an untrained series where all id values are specified, use the id values from the specified data
(if (and
(size series_context_values)
;TODO: 22655 support partial ID specification
all_id_values_defined
)
(map
(lambda
;only keep the series id values
(unzip
;create an assoc of features -> values from first row of series
(zip series_context_features (first (current_value)))
series_id_features
)
;list of lists of values
continue_series_values
)
;list of lists of values
series_context_values
)

;else pull all unique id feature combos from dataset and sample n_samples worth
(map
(lambda
;a list of feature values for each unique conjuction
;each item in the list if there were two series id features would look like so:
;(list 12345 800)
(map
(lambda
;covnvert value back to number if feature is non-string
(if (contains_index !numericNominalFeaturesMap (get series_id_features (current_index)))
(+ (current_value))
;else leave id as string
(current_value)
)
;else pull all unique id feature combos from dataset and sample n_samples worth
(map
(lambda
;a list of feature values for each unique conjuction
;each item in the list if there were two series id features would look like so:
;(list 12345 800)
(map
(lambda
;covnvert value back to number if feature is non-string
(if (contains_index !numericNominalFeaturesMap (get series_id_features (current_index)))
(+ (current_value))
;else leave id as string
(current_value)
)
(parse (current_value))
)
(parse (current_value))
)
)

;unique conjuctions of feature values for all the series features in an unparsed list (string) format
;each item in the list if there were two series id features would look like so: "(list 12345 800)"
(values
(map
(lambda
(unparse (retrieve_from_entity (current_value) series_id_features))
)
(call !AllCases)
;unique conjuctions of feature values for all the series features in an unparsed list (string) format
;each item in the list if there were two series id features would look like so: "(list 12345 800)"
(values
(map
(lambda
(unparse (retrieve_from_entity (current_value) series_id_features))
)
(true)
(call !AllCases)
)
(true)
)
)
)

(assign (assoc
initial_values_list
;when explicitly provided a list of untrained series to continue, keep that list
(if (size continue_series_values)
initial_values_list
)
)

;keeping n_samples worth of ids, sample randomly with replacement
(rand initial_values_list n_samples)
)
initial_features (append (list ".series_progress") series_id_features)
))
(assign (assoc
initial_values_list
;when explicitly provided a list of untrained series to continue, keep that list
(if (and all_id_values_defined (size series_context_values))
initial_values_list

(assign (assoc
initial_values
(map
(lambda (append (list 0.0) (current_value)) )
initial_values_list
)
))
)
;keeping n_samples worth of ids, sample randomly with replacement
(rand initial_values_list n_samples)
)
initial_features (append (list ".series_progress") series_id_features)
))

;else
(assign (assoc
initial_features (list ".series_progress")
initial_values (list (list 0.0))
initial_values
(map
(lambda (append (list 0.0) (current_value)) )
initial_values_list
)
))
)

;else
(assign (assoc
initial_features (list ".series_progress")
initial_values (list (list 0.0))
))
)
)
)
Expand Down Expand Up @@ -676,7 +673,8 @@
))

;Ensure that the user specified context features are not being derived but still being output
(if (size all_contexts)
;when forecasting, these features still need be derived
(if (and (size all_contexts) (not continue_series))
(seq
(assign (assoc
;This line ensures that some features aren't double-added to action_features later.
Expand Down
19 changes: 19 additions & 0 deletions howso/details_cases.amlg
View file
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,14 @@
;return the influential_K
(declare (assoc candidate_case_ids (trunc sorted_influence_cases influential_k)))

(if return_entity_ids_only
(conclude
(accum (assoc
output (assoc "influential_cases" candidate_case_ids)
))
)
)

;pull all the feature values for the cases
(declare (assoc
audit_data
Expand Down Expand Up @@ -244,6 +252,17 @@
))
))

(if return_entity_ids_only
(conclude
(accum (assoc
output
(assoc
"most_similar_cases" (first similar_cases)
)
))
)
)

(declare (assoc
item_retrieval_map
(zip (append
Expand Down
Loading

0 comments on commit c967371

Please sign in to comment.