Update Template sheet

[aremazeilles]

ping @alfonsotecnalia @simonePasi

Resuming activity. From what I see the code is in good shape. We would just have to update the template sheet with the appropriate info on the PI part, correct?

[simonePasi]

Yes. You mean the excel file or the readme file?

[aremazeilles]

The excel sheet. But I think it might be better we drive it, and get back to you when needed, to clarify some issues, or to validate it

[aremazeilles]

So I tried to update the template of walking complete. I placed in orange and column some comments. Might be worth discussing it a bit.

@alfonsotecnalia could you check the PI Algo spec?

Regarding each PI, if we address the first one, the other will follow I guess:

Peak_load_left: the vector contains [mean, std, step number].

I am wondering whether we could place the labels inside the PI file, as suggested in this issue. Any opinion?

On the intra-run, I think no aggregation is feasible: we cannot compress it more.
on the inter-run, this is unclear to me:
given two or three vectors provided by different runs : [ mean_1, std_1, step_number_1], [ mean_2, std_2, step_number_2], [ mean_3, std_3, step_number_3] , which computation would you suggest @simonePasi ?

[alfonsotecnalia]

I have already check the spec and include some comments.

I am wondering whether we could place the labels inside the PI file, as suggested in this issue. Any opinion?

I would say it would be good to include the label tag as the different parts of the vector have different meaning

[aremazeilles]

Resuming the action
@simonePasi , adding the label of the field within the vector is up to you. You decide. I think it is better with, but this is not mandatory.

Intra run aggregation: how to compress a PI score more

• peak_load_{left, right} = [N, N, scalar] : we could decide to limit to the mean, or leave at is (i.e not applicable)
• rms_load_{left, right}: scalar, ok
• `stance_time_{left, right} = [%, %, scalar] : same as peak
• peak_load_shoulders_{left,right} = [N, N, N, N, N, N, Nm, Nm, Nm, Nm, Nm, Nm, \#] : norm of the force? fir the torque I don't know
• rms_load_shoulders_{left, right} = [N, N, N, Nm, Nm, Nm]: like peak?

inter run aggregation: we have several captures: how to combine the pi of the different runs

• peak_load_{left, right} : we could decide to limit to the mean, or leave at is (i.e not applicable)
• rms_load_{left, right}: [0, mean, st_dev] ok
• stance_time_{left, right} : same question as for peak
• peak_load_shoulders_{left,right}: norm of the force? for the torque I don't know
• rms_load_shoulders_{left, right}: like peak?

[aremazeilles]

I updated the entries for the input files and the program command as well.

@alfonsotecnalia I left in the current format, wrench_CrutchLeft and not CrutchLeft_wrench. I have no particular preference. But the current documentation proposes that format. If you to change it please confirm it.

@simonePasi, please focus on the aggregation part, I think this is the last thing we would need to complete the metric part.

[simonePasi]

@aremazeilles I prefer to leave "standard deviation" entry in the PIs. Is it a problem? For single runs the value could be NaN

[aremazeilles]

could you identify the entry you are referring to?

[simonePasi]

Intra run aggregation:

• peak_load_{left, right} = [N] mean of the means
• rms_load_{left, right}: [N] rms of rms values
• stance_time_{left, right} = [N] mean of the means
• peak_load_shoulders_{left,right} = [N, N, N, Nm, Nm, Nm] : mean of the means both for forces and torques
• rms_load_shoulders_{left, right} = [N, N, N, Nm, Nm, Nm]: rms of the rms valuess both for forces and torques

inter run aggregation: same as intra run aggregation

[aremazeilles]

I think we still have to make some iteration here.
The intra-run aggregation is an option provided to compress a PI value, if it makes sense. For example, some a PI could be all step lenght, i.e a vector of all step lenght measured. An intra-run aggregation could consist in computing the mean (and std) of all the values of the vector. Also an inter-run aggregation (Across all repettions) woudl consist in computing mean and std of all the step lenght put together.

Let start with the first ones: peak_load_{left, right}

• one PI format : vector containing [mean, std, step number]
• intra run agg: I don't think we can do any expect focusing on mean and std
• inter run agg: meaning given score of repetition 1 [mean1, std1, step number1] and repetition 2 [mean2, std2, step number 2] ...., how can we get to a joint PI? We cna compute the mean of the mean (although mathematically I think this computation is wrong, as the the two repetition may be of different length

So my suggestion would be:

intra_run_aggregation: na (Not applicable, ie the result stays as is)
inter_run_aggregation: [mean] 

@alfonsotecnalia how do you write mean of the first entry of all vectors??

I would also suggest extending the PI format with:

type: 'vector'
label: ['mean', ' std', ' step number' ]
value: [170.73955, 27.67778, 12.00000]

[simonePasi]

I'm agree with your suggestion:
intra_run_aggregation: na for every PI

inter_run_aggregation: mean and rms values i.e.

• peak_load_{left, right} = [N] mean of the means
• rms_load_{left, right}: [N] rms of rms values
• stance_time_{left, right} = [N] mean of the means
• peak_load_shoulders_{left,right} = [N, N, N, Nm, Nm, Nm] : mean of the means both for forces and torques
• rms_load_shoulders_{left, right} = [N, N, N, Nm, Nm, Nm]: rms of the rms valuess both for forces and torques

[aremazeilles]

That would thus be:

# peak_load_{left, right}
inter_run_aggregation: [[-,0], mean] #i.e mean of all 1st entry of vector, mean of the mean
# rms_load_{left, right}
inter_run_aggregation: [[0], rms] # rms of all values
# stance_time_{left, right} 
inter_run_aggregation: [[-,0], mean] #i.e mean of all 1st entry of vector, mean of the mean

For the peak_load_shoulder and rms_load_shoulder, you expect to get a 6 entry vector, where each entry is the mean /rms value of all related vector entry, is that correct?

[simonePasi]

Correct.