Skip to content

perf(create_disrnn_dataset): group by session once instead of per-session df.query #29

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
hanhou wants to merge 1 commit into
base: main
Choose a base branch
from
Open

perf(create_disrnn_dataset): group by session once instead of per-session df.query #29

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 11 additions & 16 deletions src/aind_disrnn_utils/data_loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,26 +72,21 @@ def create_disrnn_dataset(

num_sessions = len(df_trials["ses_idx"].unique())
num_input_features = len(feature_cols)
# Pad trials to be ignored with -1
xs = np.full((max_session_length, num_sessions, num_input_features), -1)

# Load each session into xs
for dex, ses_idx in enumerate(df_trials["ses_idx"].unique()):
temp = df_trials.query("ses_idx == @ses_idx")
this_xs = temp[feature_cols].to_numpy()[:-1, :]
xs[1 : len(temp), dex, :] = this_xs # noqa E203

# Determine size of output matrix
# Output matrix has size [# trials, # sessions, # features]
num_output_features = 1
# pad trials to be ignored with -1
# Pad trials to be ignored with -1
xs = np.full((max_session_length, num_sessions, num_input_features), -1)
ys = np.full((max_session_length, num_sessions, num_output_features), -1)

# Load each session into ys
for dex, ses_idx in enumerate(df_trials["ses_idx"].unique()):
temp = df_trials.query("ses_idx == @ses_idx")
this_ys = temp[["animal_response"]].to_numpy()
ys[0 : len(temp), dex, :] = this_ys # noqa E203
# Load each session into xs/ys. Group once by session (sort=False preserves
# the first-appearance order that defines the session column index `dex`,
# matching df_trials["ses_idx"].unique()) instead of calling
# df.query("ses_idx == @ses_idx") per session. The query path re-parses the
# expression string and re-scans the frame on every call, which dominated
# dataset construction for cohorts with many sessions.
for dex, (_ses_idx, temp) in enumerate(df_trials.groupby("ses_idx", sort=False)):
xs[1 : len(temp), dex, :] = temp[feature_cols].to_numpy()[:-1, :] # noqa E203
ys[0 : len(temp), dex, :] = temp[["animal_response"]].to_numpy() # noqa E203

# Pack into a DatasetRNN object
dataset = rnn_utils.DatasetRNN(
Expand Down
Loading