The prediction of motor recovery in the acute phase of stroke is crucial for several clinical reasons: (i) informing the patient and his relatives, (ii) helping to identify the patient's endorsement (return home or rehabilitation) as well as the adaptation of the rehabilitation program to what can be expected from it. To date, an algorithm (decision tree) proposed by C. Stinear's team named PREP2 is the best predictive tool with 75% of patients well classified at 3 months. It predicts the functional recovery of the upper limb after stroke 3 months before the episode by categorizing recovery as excellent, good, limited as well as minor (poor). With two data (SAFE score, age) or three (SAFE score, Motor evoked potential, NIHSS), the prediction is effective three times out of 4. In the study the team is proposing PREP-UCV, it would like to validate this algorithm as it is with patients in the active file who are victims of stroke. The expected accuracy is 75% or more. As a secondary objective, the team would like to confirm that it find the same algorithm starting from the initial data from PREP 2 (side of the stroke, type of stroke (ischemic and / or hemorrhagic), involvement of the corticospinal tract on MRI, sex at birth ) as well as two other factors which are also very important: cognitive status (dysexecutive / aphasia / neglect), as well as the neutrophils on lymphocytes ratio.