o 2h@sddlmZddlmZddlmZddlmZddlmZddlmZddl m Z dd Z ed Gd d d eZ ed GdddeZ edGddde ZdS))backend) initializers)losses)ops) keras_export)Metric)serialization_libcCs|pt}t|}tj||d}|duritj||d}|dur+tjj|||dd}tj ||\}}t |j }t |j }||krM||t t ||d}t|t |}||}|dkri||t t d|d}t |j }|dkr}||t t d|d}||fS)Ndtypesumr reduction)axis)rfloatxget_keras_maskrcastconvert_to_tensorrloss apply_masksqueeze_or_expand_to_same_ranklenshapelistrange broadcast_to)values sample_weight reduce_fnr mask weight_ndim values_ndimr"^/var/www/html/chatgem/venv/lib/python3.10/site-packages/keras/src/metrics/reduction_metrics.pyreduce_to_samplewise_values s8     r$zkeras.metrics.Sumc<eZdZdZd fdd Zd ddZdd Zd d ZZS)SumaCompute the (weighted) sum of the given values. For example, if `values` is `[1, 3, 5, 7]` then their sum is 16. If `sample_weight` was specified as `[1, 1, 0, 0]` then the sum would be 4. This metric creates one variable, `total`. This is ultimately returned as the sum value. Args: name: (Optional) string name of the metric instance. dtype: (Optional) data type of the metric result. Example: >>> m = metrics.Sum() >>> m.update_state([1, 3, 5, 7]) >>> m.result() 16.0 >>> m = metrics.Sum() >>> m.update_state([1, 3, 5, 7], sample_weight=[1, 1, 0, 0]) >>> m.result() 4.0 r Ncs.tj||d|jdt|jdd|_dS)Nnamer r"totalr initializerr r()super__init__ add_variablerZerosr r)selfr(r  __class__r"r#r-Js z Sum.__init__cCs.t||tj|jd\}}|jt|dS)Nrr )r$rr r r) assign_add)r1rr_r"r"r# update_stateSs  zSum.update_statecCs|jddSNr)r)assignr1r"r"r# reset_stateYzSum.reset_statecCst|j|jSN)rrr)r r:r"r"r#result\r<z Sum.result)r Nr= __name__ __module__ __qualname____doc__r-r7r;r> __classcell__r"r"r2r#r&/s  r&zkeras.metrics.Meancr%)MeanaCompute the (weighted) mean of the given values. For example, if values is `[1, 3, 5, 7]` then the mean is 4. If `sample_weight` was specified as `[1, 1, 0, 0]` then the mean would be 2. This metric creates two variables, `total` and `count`. The mean value returned is simply `total` divided by `count`. Args: name: (Optional) string name of the metric instance. dtype: (Optional) data type of the metric result. Example: >>> m = Mean() >>> m.update_state([1, 3, 5, 7]) >>> m.result() 4.0 >>> m.reset_state() >>> m.update_state([1, 3, 5, 7], sample_weight=[1, 1, 0, 0]) >>> m.result() 2.0 meanNcsHtj||d|jdt|jdd|_|jdt|jdd|_dS)Nr'r"r)r*count)r,r-r.rr/r r)rGr0r2r"r#r-{s z Mean.__init__cCs|t||tj|jd\}}|jt||durt|}nt|jdkr.t|d}nd}|j tj ||jddS)Nr4rrr ) r$rrFr r)r5r rrrGr)r1rr num_samplesr"r"r#r7s   zMean.update_statecCs|jd|jddSr8)r)r9rGr:r"r"r#r;s zMean.reset_statecCst|jtj|j|jdS)Nr )r divide_no_nanr)rrGr r:r"r"r#r>sz Mean.result)rFNr=r?r"r"r2r#rE`s   rEzkeras.metrics.MeanMetricWrappercsHeZdZdZd fdd Zd fdd ZfddZed d ZZ S) MeanMetricWrapperaWrap a stateless metric function with the `Mean` metric. You could use this class to quickly build a mean metric from a function. The function needs to have the signature `fn(y_true, y_pred)` and return a per-sample loss array. `MeanMetricWrapper.result()` will return the average metric value across all samples seen so far. For example: ```python def mse(y_true, y_pred): return (y_true - y_pred) ** 2 mse_metric = MeanMetricWrapper(fn=mse) ``` Args: fn: The metric function to wrap, with signature `fn(y_true, y_pred, **kwargs)`. name: (Optional) string name of the metric instance. dtype: (Optional) data type of the metric result. **kwargs: Keyword arguments to pass on to `fn`. Nc sTtj||d||_||_|jtjvs!t|jdr&|jjtjvr(d|_dSdSdS)Nr'r3down) r,r-_fn _fn_kwargsr ALL_OBJECTShasattrr3 _direction)r1fnr(r kwargsr2r"r#r-s   zMeanMetricWrapper.__init__csVt|}|j||fi|j}|dur#|dur#tjj|||jdd}tj ||dS)Nr r )r) rrrLrMrrrr r,r7)r1y_truey_predrrrr2r"r#r7s  zMeanMetricWrapper.update_statecs8t}dt|ji}|t|ji||S)NrQ)r, get_configrserialize_keras_objectrLupdaterM)r1 base_configconfigr2r"r#rUs  zMeanMetricWrapper.get_configcCs d|vr t|}|di|S)NrQr")rdeserialize_keras_object)clsrYr"r"r# from_configs zMeanMetricWrapper.from_config)NNr=) r@rArBrCr-r7rU classmethodr\rDr"r"r2r#rJs rJN) keras.srcrrrrkeras.src.api_exportrkeras.src.metrics.metricrkeras.src.savingrr$r&rErJr"r"r"r#s      %0@