Skip to content

Table 5: Impact of dropout and SimCLR projection layer dimension on fine-tuning.

import itertools

import pandas as pd
df = pd.read_csv(
    "campaigns/ucdavis-icdm19/simclr-dropout-and-projection/campaign_summary/simclr-dropout-and-projection/summary_flowpic_dim_32.csv",
    header=[0, 1],
    index_col=[0, 1, 2],
)

df = df["acc"][["mean", "ci95"]]
df = df.T
df.columns.set_names("test_split_name", level=0, inplace=True)
df.columns.set_names("projection_layer_dim", level=1, inplace=True)
df.columns.set_names("with_dropout", level=2, inplace=True)
df = df.reorder_levels(
    ["test_split_name", "with_dropout", "projection_layer_dim"], axis=1
)

df = df[list(itertools.product(["test-script", "test-human"], [True, False], [30, 84]))]
df = df.round(2)

df.to_csv("table5_simclr_dropout_and_projectionlayer.csv")
df
test_split_name test-script test-human
with_dropout True False True False
projection_layer_dim 30 84 30 84 30 84 30 84
mean 91.81 92.02 92.18 92.54 72.12 73.31 74.69 74.35
ci95 0.38 0.36 0.31 0.33 1.37 1.04 1.13 1.38